-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathaieducation.html
More file actions
659 lines (602 loc) · 92.7 KB
/
Copy pathaieducation.html
File metadata and controls
659 lines (602 loc) · 92.7 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta name="description" content="AI in Education portfolio of Devis Saputra: four featured empirical studies and twenty-six additional AIED research prototypes.">
<title>Devis Saputra | AI in Education</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Source+Serif+4:opsz,wght@8..60,400;8..60,600;8..60,700&family=IBM+Plex+Sans:wght@400;500;600&display=swap" rel="stylesheet">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.7.2/css/all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<style>
:root{--paper:#EEF0EF;--paper-raised:#F5F6F5;--ink:#1B2430;--ink-soft:#5B6660;--line:#C9CDC5;--accent:#2F5D62;--accent-soft:#DDE6E4;--accent-deep:#244A4E;--serif:"Source Serif 4",Georgia,serif;--sans:"IBM Plex Sans",-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,sans-serif}
*{box-sizing:border-box}html{scroll-behavior:smooth}body{margin:0;background:var(--paper);color:var(--ink);font-family:var(--sans);font-size:16px;line-height:1.58}a{color:var(--accent)}h1,h2,h3{font-family:var(--serif);margin:0;line-height:1.18}
.page{display:grid;grid-template-columns:280px 1fr;max-width:1320px;margin:0 auto;min-height:100vh}.sidebar{padding:46px 32px 38px 40px;position:sticky;top:0;align-self:start;height:100vh;display:flex;flex-direction:column;justify-content:space-between}.sidebar-top h1{font-size:1.72rem}.tagline{margin:12px 0 0;color:var(--ink-soft);font-size:.98rem;max-width:22ch}nav.toc{margin-top:26px}nav.toc a{display:block;padding:6px 0 6px 12px;margin-left:-12px;border-left:2px solid transparent;color:var(--ink);text-decoration:none;font-size:.9rem;font-weight:600}nav.toc a:hover{border-left-color:var(--accent);color:var(--accent)}
.sidebar-bottom{display:grid;grid-template-columns:repeat(2,minmax(0,1fr));column-gap:12px;row-gap:9px;font-size:.84rem;color:var(--ink-soft)}.sidebar-bottom .contact-label{grid-column:1/-1;color:var(--ink);font-weight:600}.sidebar-bottom a{display:flex;align-items:center;gap:7px;color:var(--accent);text-decoration:none;white-space:nowrap}.sidebar-bottom i{width:18px;text-align:center}.sidebar-bottom .scholar{grid-column:1/-1}
.content{padding:48px 48px 80px 34px;border-left:1px solid var(--line)}section{max-width:940px;margin-bottom:56px}.back-link{display:inline-flex;gap:8px;align-items:center;margin-bottom:25px;color:var(--ink-soft);text-decoration:none;font-size:.9rem}
.hero{margin-bottom:48px}.kicker{font-size:.78rem;text-transform:uppercase;letter-spacing:.13em;color:var(--accent);font-weight:600;margin-bottom:10px}.hero h2{font-size:clamp(2rem,3.6vw,2.8rem)}.hero .lede{margin:16px 0 0;max-width:840px;font-family:var(--serif);font-size:1.13rem;color:#303B42}
.portfolio-nav{display:grid;grid-template-columns:repeat(4,1fr);margin-top:28px;border:1px solid #D7DAD5;border-radius:7px;overflow:hidden}.portfolio-nav a{display:flex;align-items:center;justify-content:center;min-height:48px;padding:9px 8px;background:#F2F3F2;color:#222B33;text-decoration:none;text-align:center;font-size:.88rem;font-weight:600;border-right:1px solid #D7DAD5}.portfolio-nav a:last-child{border-right:none}.portfolio-nav a.active{background:var(--accent);color:#fff}
.eyebrow-rule{display:flex;align-items:center;gap:14px;margin-bottom:18px}.eyebrow-rule h2{font-size:1.45rem}.eyebrow-rule::after{content:"";flex:1;height:1px;background:var(--line)}
.contact-row{display:flex;flex-wrap:wrap;gap:10px}.contact-row a{padding:8px 13px;border:1px solid var(--line);border-radius:5px;background:var(--paper-raised);color:var(--ink);text-decoration:none}
footer{max-width:940px;margin-top:54px;padding-top:18px;border-top:1px solid var(--line);color:var(--ink-soft);font-size:.78rem}
@media(max-width:900px){.page{grid-template-columns:1fr}.sidebar{position:static;height:auto;padding:34px 24px 26px;border-bottom:1px solid var(--line);gap:26px}nav.toc{display:none}.content{padding:34px 24px 58px;border-left:none}}
@media(max-width:700px){.portfolio-nav{grid-template-columns:1fr}.portfolio-nav a{border-right:none;border-bottom:1px solid #D7DAD5}}
.project-index{display:flex;flex-wrap:wrap;gap:7px;margin-top:22px}
.project-index a{padding:5px 9px;border:1px solid var(--line);border-radius:5px;background:var(--paper-raised);text-decoration:none;font-size:.78rem;color:#39444A}
.project-index a:hover,.project-index a:focus-visible{border-color:var(--accent);color:var(--accent)}
.portfolio-detail .repo-visual img{aspect-ratio:16/9;height:auto}.repo-copy{line-height:1.65}.repo-links a{min-height:40px}
</style>
<link rel="stylesheet" href="assets/site.css?v=20260927-aied-empirical-3">
</head>
<body class="portfolio-detail">
<div class="page">
<aside class="sidebar">
<div class="sidebar-top">
<h1>Devis Saputra</h1>
<p class="tagline">Learning Designer | AIEd</p>
<nav class="toc" aria-label="AI in Education page navigation">
<a href="index.html">← Main Portfolio</a>
<a href="#overview">Overview</a>
<a href="#research-bundles">Empirical Studies <i class="fa-brands fa-github" aria-hidden="true"></i></a>
<a href="#repositories">Repositories</a>
<a href="#contact">Contact</a>
</nav>
</div>
<div class="sidebar-bottom" aria-label="Contact links">
<div class="contact-label">Links</div>
<a href="mailto:devis.saputra@gmail.com"><i class="fa-solid fa-envelope"></i><span>Email</span></a>
<a href="https://linkedin.com/in/devissaputra" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-linkedin"></i><span>LinkedIn</span></a>
<a href="https://www.xing.com/profile/Devis_Saputra" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-xing"></i><span>Xing</span></a>
<a href="https://github.com/devissaputra/" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i><span>GitHub</span></a>
<a href="https://orcid.org/0000-0002-7133-9410" target="_blank" rel="noopener noreferrer"><i class="ai ai-orcid"></i><span>ORCID</span></a>
<a class="scholar" href="https://scholar.google.com/citations?user=EB5_ZRMAAAAJ&hl=en&oi=ao" target="_blank" rel="noopener noreferrer"><i class="ai ai-google-scholar"></i><span>Google Scholar</span></a>
</div>
</aside>
<main class="content">
<a class="back-link" href="index.html"><i class="fa-solid fa-arrow-left"></i> Back to main portfolio</a>
<div class="portfolio-nav" aria-label="Portfolio areas">
<a href="research.html">Research</a>
<a href="learningdevelopment.html">Learning & Development</a>
<a href="aiengineering.html">AI Engineering</a>
<a class="active" href="aieducation.html" aria-current="page">AI in Education</a>
</div>
<section class="hero" id="overview">
<h2>Portfolio in AI in Education</h2>
<p class="lede">This portfolio connects learning design with machine learning, learning analytics and responsible AI. Empirical studies examine knowledge tracing, observational learning analytics, retrieval and model auditing using external datasets, while the repository section presents research prototypes for assessment, learner support, curriculum analysis and educational decision support.</p>
<div class="project-index" aria-label="Page content">
<a href="#research-bundles">Empirical Studies</a>
<a href="#repositories">Repositories</a>
</div>
</section>
<section class="repo-section" id="research-bundles">
<div class="eyebrow-rule bundle-heading"><h2>Empirical Studies</h2></div>
<p class="repo-section-intro">These four studies use external educational datasets to answer four different questions: how learner history improves next-response prediction, how early assessment submission relates to later outcomes, whether a wrong answer helps evidence retrieval, and how an enrollment-time dropout model behaves under a responsible audit. Each case is presented in the same order: research question, data, analytical design, result and interpretation boundary.</p>
<article class="repo-project research-bundle" id="knowledge_tracing_benchmark">
<header class="repo-project-heading"><div class="repo-kicker">Empirical Study · AI in Education</div>
<h3>Knowledge Tracing Benchmark on ASSISTments 2009</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/knowledge_tracing_benchmark/review_overview.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open knowledge tracing study overview at full size"><img loading="lazy" src="assets/aied/knowledge_tracing_benchmark/review_overview.svg?v=20260927-aied-empirical-3" alt="Scientific overview of the ASSISTments 2009 knowledge tracing benchmark, including study question, learner history models, held-out results and interpretation limits."></a><figcaption>Full study story: question, data, design, result and interpretation limit</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/knowledge_tracing_benchmark/review_calculations.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open knowledge tracing data processing pipeline at full size"><img loading="lazy" src="assets/aied/knowledge_tracing_benchmark/review_calculations.svg?v=20260927-aied-empirical-3" alt="Scientific data processing and evaluation pipeline for the ASSISTments 2009 knowledge tracing benchmark, from learner sequences and data splits through model fitting and held-out evaluation."></a><figcaption>Data processing and analysis workflow</figcaption></figure>
</div>
<div>
<div class="repo-copy">
<p>This study asks a straightforward learner-modeling question: when a model can use a learner's previous attempts, how much better can it predict the learner's next response? The benchmark uses 4,148 ASSISTments 2009 learners and 274,331 interactions, with a 70/15/15 split made by learner rather than by attempt. It compares simple population and skill priors with Bayesian Knowledge Tracing, a PFA-style logistic model and a compact GRU. For every stateful model, the prediction is produced before the current answer is revealed, preventing target leakage.</p>
<p>On 623 held-out learners, the seed-42 GRU reaches ROC-AUC 0.7470 and Brier score 0.1800, while PFA reaches 0.6982 and 0.1976. Across three GRU seeds, mean ROC-AUC is 0.7471 with a sample standard deviation of 0.0008. The result is therefore about predictive performance on this historical benchmark: it shows that sequence information adds useful signal here, but it does not turn the model probability into a direct measure of knowledge or show that using the model would improve teaching or learning.</p>
</div>
<div class="repo-evidence" aria-label="Knowledge tracing evidence snapshot">
<div class="repo-evidence-item"><span class="repo-evidence-value">4,148</span><span class="repo-evidence-label">learners</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">274,331</span><span class="repo-evidence-label">interactions</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">623</span><span class="repo-evidence-label">test learners</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">0.7471</span><span class="repo-evidence-label">GRU mean ROC-AUC</span></div>
</div>
<div class="repo-data-note"><strong>Data and provenance:</strong> ASSISTments 2009 is the scientific source. The executable adapter uses the public Atomi sequence representation at pinned revision <code>c72a664…</code>, verifies its SHA-256 and learner-row integrity, and keeps the raw learner data outside the repository.</div>
<div class="repo-links">
<a href="https://github.com/devissaputra/knowledge_tracing_benchmark/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a>
<a href="https://github.com/devissaputra/knowledge_tracing_benchmark" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/knowledge_tracing_benchmark/blob/main/paper/results.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-chart-line"></i> Results</a>
<a href="https://github.com/devissaputra/knowledge_tracing_benchmark/blob/main/paper/paper.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-pen"></i> Paper</a>
<a href="https://github.com/devissaputra/knowledge_tracing_benchmark/blob/main/DATA.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-database"></i> Data & provenance</a>
<a href="https://github.com/devissaputra/knowledge_tracing_benchmark/blob/main/docs/research_protocol.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-flask"></i> Protocol</a>
</div>
</div>
</div>
</article>
<article class="repo-project research-bundle" id="causal_learning_analytics">
<header class="repo-project-heading"><div class="repo-kicker">Empirical Study · AI in Education</div>
<h3>Early Assessment Submission and Later Outcomes in OULAD</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/causal_learning_analytics/review_overview.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open OULAD landmark study overview at full size"><img loading="lazy" src="assets/aied/causal_learning_analytics/review_overview.svg?v=20260927-aied-empirical-3" alt="Scientific overview of the OULAD day 30 landmark study, including eligibility, exposure, outcome timing, adjusted contrast and causal interpretation limits."></a><figcaption>Full study story: question, data, design, result and interpretation limit</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/causal_learning_analytics/review_calculations.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open OULAD data processing and analysis pipeline at full size"><img loading="lazy" src="assets/aied/causal_learning_analytics/review_calculations.svg?v=20260927-aied-empirical-3" alt="Scientific processing pipeline for the OULAD day 30 landmark analysis, from cohort definition and exposure through weighting, diagnostics, bootstrap uncertainty and sensitivity analysis."></a><figcaption>Data processing and analysis workflow</figcaption></figure>
</div>
<div>
<div class="repo-copy">
<p>This study examines whether submitting at least one qualifying non-banked assessment by day 30 is associated with a more favorable final outcome among learners who are still eligible at that point. Using the OULAD CCC 2014J presentation, the day-30 landmark is fixed before exposure and outcome are defined. The analysis starts with 1,983 eligible learners and retains 1,820 complete supported cases: 1,624 had submitted by the landmark and 196 had not. Pre-exposure covariates are then used to construct inverse-probability weights for the comparison.</p>
<p>The normalized Hájek contrast is 0.3861, with a full-refit bootstrap interval of approximately 0.3242 to 0.4447; the overlap-weighted sensitivity estimate is 0.3261. Those numbers are reported together with the less comfortable diagnostics: extreme weights, an effective sample size of about 509.5 and residual imbalance remain. The study therefore supports an adjusted observational association, not the claim that making students submit earlier would cause better outcomes.</p>
</div>
<div class="repo-evidence" aria-label="OULAD landmark evidence snapshot">
<div class="repo-evidence-item"><span class="repo-evidence-value">1,820</span><span class="repo-evidence-label">analysis cases</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">1,624</span><span class="repo-evidence-label">exposed</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">196</span><span class="repo-evidence-label">control</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">0.3861</span><span class="repo-evidence-label">Hájek contrast</span></div>
</div>
<div class="repo-data-note"><strong>Data and provenance:</strong> Open University Learning Analytics Dataset, OULAD / UCI 349, using the frozen CCC 2014J day 30 landmark design. The recorded analysis files preserve the cohort definition, weighting diagnostics, uncertainty procedure and source fingerprint.</div>
<div class="repo-links">
<a href="https://github.com/devissaputra/causal_learning_analytics/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a>
<a href="https://github.com/devissaputra/causal_learning_analytics" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/causal_learning_analytics/blob/main/paper/results.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-chart-line"></i> Results</a>
<a href="https://github.com/devissaputra/causal_learning_analytics/blob/main/paper/paper.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-pen"></i> Paper</a>
<a href="https://github.com/devissaputra/causal_learning_analytics/blob/main/DATA.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-database"></i> Data & provenance</a>
<a href="https://github.com/devissaputra/causal_learning_analytics/blob/main/docs/research_protocol.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-flask"></i> Protocol</a>
</div>
</div>
</div>
</article>
<article class="repo-project research-bundle" id="misconception_aware_rag">
<header class="repo-project-heading"><div class="repo-kicker">Empirical Study · AI in Education</div>
<h3>Wrong Answer Conditioned Retrieval on SciQ</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/misconception_aware_rag/review_overview.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open misconception aware retrieval study overview at full size"><img loading="lazy" src="assets/aied/misconception_aware_rag/review_overview.svg?v=20260927-aied-empirical-3" alt="Scientific overview of the SciQ retrieval study, comparing question-only retrieval with observed wrong-answer expansion and a shuffled distractor control."></a><figcaption>Full study story: question, data, design, result and interpretation limit</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/misconception_aware_rag/review_calculations.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open misconception aware retrieval pipeline at full size"><img loading="lazy" src="assets/aied/misconception_aware_rag/review_calculations.svg?v=20260927-aied-empirical-3" alt="Scientific retrieval evaluation pipeline for the SciQ study, from supported questions and query construction through BM25 retrieval, ranking metrics and paired question-block comparisons."></a><figcaption>Data processing and analysis workflow</figcaption></figure>
</div>
<div>
<div class="repo-copy">
<p>This study tests one specific idea behind a misconception-aware tutor: does the wrong answer itself contain useful information for retrieving the evidence that should support a response? From the SciQ test split, 884 supported questions provide 884 support passages and 2,652 question–wrong-answer proxy cases. A fixed BM25 setup compares three queries for the same problem: the question alone, the question plus its observed wrong answer, and the question plus a shuffled wrong answer. The shuffled condition acts as a negative control for the simple effect of adding more lexical material.</p>
<p>Question-only retrieval achieves MRR 0.9472, compared with 0.9433 for observed wrong-answer expansion and 0.9447 for the shuffled control. The observed wrong-answer condition is 0.0039 below the baseline, and its difference from the shuffled control is small with a 95% interval spanning zero. In this benchmark, the wrong answer does not provide a distinct retrieval advantage. That conclusion is deliberately narrow: SciQ distractors are answer options, not validated learner misconceptions, and the experiment evaluates retrieval rather than learning gains from a tutor.</p>
</div>
<div class="repo-evidence" aria-label="Misconception aware retrieval evidence snapshot">
<div class="repo-evidence-item"><span class="repo-evidence-value">884</span><span class="repo-evidence-label">eligible questions</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">2,652</span><span class="repo-evidence-label">wrong-answer cases</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">0.9472</span><span class="repo-evidence-label">baseline MRR</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">−0.0039</span><span class="repo-evidence-label">wrong-answer Δ MRR</span></div>
</div>
<div class="repo-data-note"><strong>Data and provenance:</strong> SciQ test split from Allen Institute for AI, pinned to revision <code>2c94ad3…</code> with byte-level SHA-256 verification. The empirical corpus uses 884 supported questions, and the shuffled condition is a deterministic lexical-expansion control.</div>
<div class="repo-links">
<a href="https://github.com/devissaputra/misconception_aware_rag/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a>
<a href="https://github.com/devissaputra/misconception_aware_rag" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/misconception_aware_rag/blob/main/paper/results.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-chart-line"></i> Results</a>
<a href="https://github.com/devissaputra/misconception_aware_rag/blob/main/paper/paper.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-pen"></i> Paper</a>
<a href="https://github.com/devissaputra/misconception_aware_rag/blob/main/DATA.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-database"></i> Data & provenance</a>
<a href="https://github.com/devissaputra/misconception_aware_rag/blob/main/docs/research_protocol.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-flask"></i> Protocol</a>
</div>
</div>
</div>
</article>
<article class="repo-project research-bundle" id="responsible_aied_evaluation">
<header class="repo-project-heading"><div class="repo-kicker">Empirical Study · AI in Education</div>
<h3>Enrollment-Time Dropout Risk Model Audit on UCI 697</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/responsible_aied_evaluation/review_overview.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open responsible AIED audit overview at full size"><img loading="lazy" src="assets/aied/responsible_aied_evaluation/review_overview.svg?v=20260927-aied-empirical-3" alt="Scientific overview of the UCI 697 responsible AIED audit, including enrollment-time prediction boundaries, model performance, subgroup support and interpretation limits."></a><figcaption>Full study story: question, data, design, result and interpretation limit</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/responsible_aied_evaluation/review_calculations.svg?v=20260927-aied-empirical-3" target="_blank" rel="noopener noreferrer" aria-label="Open responsible AIED audit pipeline at full size"><img loading="lazy" src="assets/aied/responsible_aied_evaluation/review_calculations.svg?v=20260927-aied-empirical-3" alt="Scientific data processing and audit pipeline for UCI 697, from prediction-time feature selection through model evaluation, subgroup audit, uncertainty checks and governance limitations."></a><figcaption>Data processing and analysis workflow</figcaption></figure>
</div>
<div>
<div class="repo-copy">
<p>This study audits an enrollment-time dropout-risk model rather than presenting it as a deployment-ready predictor. UCI dataset 697 contains 4,424 historical student records. Variables that describe first- or second-semester performance are removed so the prediction boundary stays at enrollment, while protected attributes are kept out of the model and used only for auditing. The resulting model uses 20 predictors and is examined for discrimination, calibration, subgroup error patterns, threshold-dependent allocation and sensitivity to alternative data splits.</p>
<p>On the primary 885-record holdout, ROC-AUC is 0.8316 and Brier score is 0.1447. The gender audit has enough support under the study's stated rules, while international status, intersectional groups and special-needs comparisons do not; those cases are marked not evaluable rather than being interpreted as zero disparity. Repeated splits and bootstrap analyses add uncertainty information, but the evidence still belongs to one historical Portuguese higher-education dataset. The result is an audit record, not a fairness certificate or permission to deploy the model.</p>
</div>
<div class="repo-evidence" aria-label="Responsible AIED evaluation evidence snapshot">
<div class="repo-evidence-item"><span class="repo-evidence-value">4,424</span><span class="repo-evidence-label">records</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">20</span><span class="repo-evidence-label">retained predictors</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">0.8316</span><span class="repo-evidence-label">ROC-AUC</span></div>
<div class="repo-evidence-item"><span class="repo-evidence-value">0.1447</span><span class="repo-evidence-label">Brier score</span></div>
</div>
<div class="repo-data-note"><strong>Data and provenance:</strong> UCI Machine Learning Repository dataset 697, <em>Predict Students' Dropout and Academic Success</em>, DOI 10.24432/C5MC89. Each empirical run records a SHA-256 fingerprint of the normalized source table and target.</div>
<div class="repo-links">
<a href="https://github.com/devissaputra/responsible_aied_evaluation/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a>
<a href="https://github.com/devissaputra/responsible_aied_evaluation" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/responsible_aied_evaluation/blob/main/paper/results.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-chart-line"></i> Results</a>
<a href="https://github.com/devissaputra/responsible_aied_evaluation/blob/main/paper/paper.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-pen"></i> Paper</a>
<a href="https://github.com/devissaputra/responsible_aied_evaluation/blob/main/DATA.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-database"></i> Data & provenance</a>
<a href="https://github.com/devissaputra/responsible_aied_evaluation/blob/main/docs/research_protocol.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-flask"></i> Protocol</a>
</div>
</div>
</div>
</article>
</section>
<section class="repo-section" id="repositories">
<div class="eyebrow-rule"><h2>Repositories</h2></div>
<p class="repo-section-intro">The remaining 26 projects are research prototypes and technical studies. Their summaries distinguish implemented methods from claims that still require empirical validation.</p>
<article class="repo-project" id="genai_learning_observatory">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 05</div>
<h3>GenAI Learning Observatory</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/genai_learning_observatory/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/genai_learning_observatory/review_overview.svg?v=20260926-evidence-review" alt="How do learners check and revise AI assistance? Synthetic prompt-event records grouped by learner and session."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/genai_learning_observatory/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/genai_learning_observatory/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Agency index = .30V + .25R + .25F + .20(1-A)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>GenAI Learning Observatory converts synthetic AI interaction logs into session summaries of verification, revision, reflection and answer adoption. Intent labels are defined by explicit lexical rules, and the agency index is a weighted heuristic rather than a psychological measure. The implementation is useful for testing instrumentation and analysis procedures before collecting consented learner data.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/genai_learning_observatory/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/genai_learning_observatory" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/genai_learning_observatory/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="teacher_ai_assessment">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 06</div>
<h3>Teacher and AI Assessment Studio</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/teacher_ai_assessment/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/teacher_ai_assessment/review_overview.svg?v=20260926-evidence-review" alt="Which human–AI scoring disagreements need review? Synthetic paired human and AI ordinal scores with uncertainty values."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/teacher_ai_assessment/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/teacher_ai_assessment/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Review = |human-AI| > 1 OR uncertainty ≥ .28."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Teacher AI Assessment Studio compares synthetic human and AI assessment scores and creates a review queue from explicit disagreement and uncertainty thresholds. It reports exact agreement, agreement within one score point, quadratic kappa and the proportion routed for review. These outputs test routing behavior; they do not establish the correctness of an AI assessor or validate the thresholds for real students.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/teacher_ai_assessment/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/teacher_ai_assessment" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/teacher_ai_assessment/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="privacy_preserving_learning_analytics">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 07</div>
<h3>Privacy Preserving Learning Analytics</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/privacy_preserving_learning_analytics/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/privacy_preserving_learning_analytics/review_overview.svg?v=20260926-evidence-review" alt="What changes when training remains within data silos? Synthetic client shards and a separate synthetic test set."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/privacy_preserving_learning_analytics/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/privacy_preserving_learning_analytics/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: g = Xᵀ(sigmoid(Xw)-y)/n; w_next = w - learning_rate×mean(client gradients)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Privacy Preserving Learning Analytics demonstrates federated learning with synthetic learner records stored in local shards. Clients share clipped gradients, optionally with added noise, while a separate synthetic test set measures the effect on predictive utility. The implementation exposes the update rule and client weighting. Formal privacy accounting, secure aggregation and adversarial evaluation remain outside the current scope.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/privacy_preserving_learning_analytics/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/privacy_preserving_learning_analytics" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/privacy_preserving_learning_analytics/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="multimodal_self_regulation_lab">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 08</div>
<h3>Multimodal Self Regulation Lab</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/multimodal_self_regulation_lab/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/multimodal_self_regulation_lab/review_overview.svg?v=20260926-evidence-review" alt="Does combining modalities improve synthetic prediction? Simulated interaction, attention-like and self-report features with missingness and reliability values."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/multimodal_self_regulation_lab/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/multimodal_self_regulation_lab/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Fusion = (interaction_rank + reliabilityA×attention_rank + reliabilityR×report_rank)/(1+reliabilityA+reliabilityR)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Multimodal Self Regulation Lab compares synthetic interaction, attention proxy and self report signals using one common held out partition. The classifier pipeline fits imputation and scaling only on training rows, correcting an earlier evaluation design that reused information from the full sample. A separate rank fusion comparator explores reliability weighting. Because both the signals and outcomes are simulated, the results do not validate measurement of self regulation or attention in real learners.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/multimodal_self_regulation_lab/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/multimodal_self_regulation_lab" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/multimodal_self_regulation_lab/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="learning_design_process_mining">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 09</div>
<h3>Learning Design Process Mining</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learning_design_process_mining/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learning_design_process_mining/review_overview.svg?v=20260926-evidence-review" alt="Where does instructional-design work repeat or branch? Synthetic timestamped design-event logs, grouped into cases."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learning_design_process_mining/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learning_design_process_mining/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: P(next|stage) = transition count / outgoing count; H(stage) = -sum p log2 p."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Learning Design Process Mining reconstructs synthetic instructional design event traces and reports common process variants, stage transition shares, entropy and a defined repetition rate. The measures make workflow patterns visible without treating every revision as inefficiency. Meaningful conclusions would require authentic design logs and a defensible interpretation of the stage labels.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/learning_design_process_mining/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/learning_design_process_mining" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/learning_design_process_mining/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="hybrid_intelligence_lab">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 10</div>
<h3>Hybrid Intelligence Lab</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/hybrid_intelligence_lab/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/hybrid_intelligence_lab/review_overview.svg?v=20260926-evidence-review" alt="When does confidence-based routing help? Synthetic human and AI decisions; confidence is generated partly from correctness."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/hybrid_intelligence_lab/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/hybrid_intelligence_lab/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Complementarity = mean(human_correct XOR AI_correct)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Hybrid Intelligence Lab compares human decisions, AI decisions and confidence routed combinations while retaining the routing reason for every case. The synthetic generator deliberately links confidence with correctness, so strong combined performance partly reflects that assumption. Threshold sweeps show how often the policy defaults to human judgment and under which simulated conditions complementary errors improve combined performance.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/hybrid_intelligence_lab/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/hybrid_intelligence_lab" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/hybrid_intelligence_lab/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="explanation_faithfulness_aied">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 11</div>
<h3>Explanation Faithfulness for AIED</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/explanation_faithfulness_aied/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/explanation_faithfulness_aied/review_overview.svg?v=20260926-evidence-review" alt="Do chosen features retain and explain a prediction? Synthetic original, feature-removed and feature-only probabilities plus feature sets."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/explanation_faithfulness_aied/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/explanation_faithfulness_aied/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Comprehensiveness = max(0,p-p_without); sufficiency gap = max(0,p-p_only)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Explanation Faithfulness for AIED computes explanation diagnostics from probability changes and feature overlap on synthetic cases. Comprehensiveness, sufficiency gap and stability are reported separately, while the combined index is identified as a weighted heuristic. The current software accepts supplied perturbation outputs. A substantive empirical evaluation would require actual model interventions and suitable random feature controls.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/explanation_faithfulness_aied/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/explanation_faithfulness_aied" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/explanation_faithfulness_aied/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="collaborative_reasoning_analytics">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 12</div>
<h3>Collaborative Reasoning Analytics</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/collaborative_reasoning_analytics/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/collaborative_reasoning_analytics/review_overview.svg?v=20260926-evidence-review" alt="How can dialogue participation and uptake be inspected? Synthetic ordered dialogue turns and speaker identities."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/collaborative_reasoning_analytics/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/collaborative_reasoning_analytics/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Uptake proxy = |tokens(previous) ∩ tokens(current)| / |union|."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Collaborative Reasoning Analytics combines explicit dialogue move labels, lexical overlap between adjacent turns and speaker participation balance on synthetic conversations. These quantities can support inspection of a coding procedure, but shared vocabulary does not establish conceptual uptake and balanced participation does not establish equitable reasoning. Human annotation and validation are therefore required before substantive interpretation.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/collaborative_reasoning_analytics/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/collaborative_reasoning_analytics" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/collaborative_reasoning_analytics/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="cognitive_offloading_analytics">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 13</div>
<h3>Cognitive Offloading Analytics</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/cognitive_offloading_analytics/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/cognitive_offloading_analytics/review_overview.svg?v=20260926-evidence-review" alt="How sensitive is an offloading proxy to its weights? Synthetic learner summaries of copying, revision, verification, recall and confidence change."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/cognitive_offloading_analytics/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/cognitive_offloading_analytics/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Normalized score = (weighted score - cohort min)/(cohort max - cohort min + 1e-9)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Cognitive Offloading Analytics combines copying similarity, revision, verification, delayed recall and confidence change into an explicit offloading proxy. The chosen weights are visible and alternative weighting scenarios are tested. Scores are normalized within each cohort and the bands are heuristic, so the results support sensitivity analysis rather than diagnosis of individual dependence on AI.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/cognitive_offloading_analytics/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/cognitive_offloading_analytics" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/cognitive_offloading_analytics/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="classroom_discourse_intelligence">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 14</div>
<h3>Classroom Discourse Intelligence</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/classroom_discourse_intelligence/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/classroom_discourse_intelligence/review_overview.svg?v=20260926-evidence-review" alt="Can teacher talk moves be classified across transcripts? TalkMoves adapter and protocol; committed empirical metrics currently require regeneration."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/classroom_discourse_intelligence/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/classroom_discourse_intelligence/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Macro-F1 = mean(classwise 2×precision×recall/(precision+recall))."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Classroom Discourse Intelligence evaluates teacher talk move classification on the TalkMoves corpus. A majority baseline is compared with logistic regression using TF-IDF features after duplicate transcript copies are removed and matching transcript text is grouped before the split. The final dataset contains 565 transcript groups and 175,129 labeled teacher utterances. On 113 held out groups, logistic regression reaches macro-F1 0.5198 compared with 0.1152 for the majority baseline, while accuracy remains similar. The result concerns coded discourse moves and does not measure teaching quality.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/classroom_discourse_intelligence/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/classroom_discourse_intelligence" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/classroom_discourse_intelligence/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="adaptive_socratic_tutor">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 15</div>
<h3>Adaptive Socratic Tutor</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/adaptive_socratic_tutor/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/adaptive_socratic_tutor/review_overview.svg?v=20260926-evidence-review" alt="How much help should a transparent tutoring policy offer? Supplied mastery and attempt counts; small synthetic examples."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/adaptive_socratic_tutor/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/adaptive_socratic_tutor/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Hint level ∈ {0,1,2,3}; lower mastery or more attempts can increase support."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Adaptive Socratic Tutor applies a transparent support policy to supplied mastery estimates and attempt counts. The rules move from a focused question toward a worked substep while keeping the complete answer outside the main response policy. Synthetic cases make the decision boundaries inspectable, but the package neither estimates mastery itself nor demonstrates an effect on learning.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/adaptive_socratic_tutor/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/adaptive_socratic_tutor" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/adaptive_socratic_tutor/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="instructor_insight_engine">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 16</div>
<h3>Instructor Insight Engine</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/instructor_insight_engine/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/instructor_insight_engine/review_overview.svg?v=20260926-evidence-review" alt="Which learner summaries deserve a closer look? Synthetic events containing learner IDs, minutes, attempts and correct counts."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/instructor_insight_engine/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/instructor_insight_engine/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Learner accuracy = total correct / total attempts; flag if accuracy < .5 and attempts ≥ 3."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Instructor Insight Engine converts synthetic activity events into learner summaries and simple flags for instructor review. Accuracy is calculated from aggregated correct and attempted responses, and the low accuracy flag requires at least three attempts. Class level summaries are also reported. Thresholds remain unvalidated, time spent is not treated as learning, and sparse records require human interpretation.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/instructor_insight_engine/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/instructor_insight_engine" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/instructor_insight_engine/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="learner_state_sequence_model">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 17</div>
<h3>Learner State Sequence Model</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learner_state_sequence_model/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learner_state_sequence_model/review_overview.svg?v=20260926-evidence-review" alt="How often do observed learner states persist or change? A supplied sequence of state labels, not inferred psychological states."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learner_state_sequence_model/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learner_state_sequence_model/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: P(j|i) = count(i→j)/sum_j count(i→j); persistence = self-transitions/(n-1)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Learner State Sequence Model summarizes supplied learner state sequences through transition probabilities, entropy of outgoing states and persistence. Probabilities are normalized only where transitions are observed, and insufficient sequences return an undefined persistence value rather than an invented estimate. The method is a descriptive sequence baseline; the state labels require independent justification and are not interpreted as inferred mental states.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/learner_state_sequence_model/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/learner_state_sequence_model" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/learner_state_sequence_model/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="multimodal_learning_analytics">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 18</div>
<h3>Multimodal Learning Analytics</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/multimodal_learning_analytics/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/multimodal_learning_analytics/review_overview.svg?v=20260926-evidence-review" alt="How can asynchronous modalities be aligned without mixing learners? Synthetic timestamped speech, gaze and click events with learner/session identifiers."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/multimodal_learning_analytics/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/multimodal_learning_analytics/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Window index = floor(timestamp/window width); speech/gaze = observed means; clicks = sum."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Multimodal Learning Analytics groups supplied multimodal observations by learner, session and fixed time window before summarizing available speech, gaze and click fields. Missing modalities remain explicit, and the interface prevents accidental mixing of identified learner sessions. The output is a data processing baseline and does not infer attention, emotion or learning.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/multimodal_learning_analytics/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/multimodal_learning_analytics" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/multimodal_learning_analytics/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="constructive_alignment_auditor">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 19</div>
<h3>Constructive Alignment Auditor</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/constructive_alignment_auditor/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/constructive_alignment_auditor/review_overview.svg?v=20260926-evidence-review" alt="Where might objectives, activities and assessments be misaligned? Supplied instructional statements and a transparent Bloom-verb lexicon."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/constructive_alignment_auditor/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/constructive_alignment_auditor/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Content overlap = shared content tokens / union of content tokens."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Constructive Alignment Auditor compares learning objectives, activities and assessments using an explicit Bloom verb lexicon and overlap among content terms. The evidence behind each inferred level is reported and uncertain or mismatched components are flagged for review. The lexical rules are intentionally inspectable, and the output is intended to support instructional judgment rather than function as a validated alignment score.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/constructive_alignment_auditor/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/constructive_alignment_auditor" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/constructive_alignment_auditor/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="assessment_design_lab">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 20</div>
<h3>Assessment Design Lab</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/assessment_design_lab/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/assessment_design_lab/review_overview.svg?v=20260926-evidence-review" alt="Which assessment items warrant review? Synthetic binary response matrices; optional criterion scores."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/assessment_design_lab/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/assessment_design_lab/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Alpha = k/(k-1) × (1 - sum(item variances)/variance(total score))."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Assessment Design Lab computes item difficulty, discrimination based on extreme groups and internal consistency diagnostics from synthetic response matrices. Rest scores exclude the focal item when comparison groups are formed, and undefined quantities remain explicit when the data cannot support an estimate. Flagged items are presented for review rather than treating alpha or a threshold as proof of assessment validity.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/assessment_design_lab/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/assessment_design_lab" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/assessment_design_lab/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="learning_experiment_platform">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 21</div>
<h3>Learning Experiment Platform</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learning_experiment_platform/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learning_experiment_platform/review_overview.svg?v=20260926-evidence-review" alt="How should a small learning experiment report its outcomes? Synthetic assignment records with baseline, follow-up and missing-outcome indicators."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learning_experiment_platform/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learning_experiment_platform/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Cohen d = (mean treatment - mean control)/pooled sample SD."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Learning Experiment Platform stores assignment, baseline measures, follow up outcomes and attrition in one record. It reports group differences, effect sizes based on pooled standard deviation and bootstrap intervals while flagging sparse groups and baseline imbalance. Missing outcomes are handled as complete cases grouped by assignment, so the documentation limits causal interpretation and does not describe the analysis as a complete intention-to-treat analysis.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/learning_experiment_platform/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/learning_experiment_platform" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/learning_experiment_platform/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="lesson_design_agent">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 22</div>
<h3>Lesson Design Agent</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/lesson_design_agent/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/lesson_design_agent/review_overview.svg?v=20260926-evidence-review" alt="How can lesson-planning assumptions be made explicit? Supplied objective, duration, delivery mode and accessibility needs."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/lesson_design_agent/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/lesson_design_agent/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Allocated phase minutes sum to the requested lesson duration."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Lesson Design Agent converts a supplied objective and teaching constraints into a structured lesson plan using transparent rules. It allocates time across lesson phases, proposes assessment and accessibility prompts and flags missing or ambiguous planning inputs. The generated activities remain proposals that require contextual review by an educator.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/lesson_design_agent/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/lesson_design_agent" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/lesson_design_agent/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="learner_agency_simulator">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 23</div>
<h3>Learner Agency Simulator</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learner_agency_simulator/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learner_agency_simulator/review_overview.svg?v=20260926-evidence-review" alt="What do support policies imply under stated transition assumptions? Synthetic state-transition matrices, learner preferences and support policies."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/learner_agency_simulator/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/learner_agency_simulator/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: State occupancy = observations in state / all recorded states."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Learner Agency Simulator compares support policies while allowing synthetic learners to request, accept or decline assistance. It records state occupancy, intervention burden, recovery and time to first mastery across seeded trajectories and sensitivity scenarios. The results describe the consequences of the assumed transition rules and do not establish that a policy improves agency or learning outside the simulation.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/learner_agency_simulator/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/learner_agency_simulator" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/learner_agency_simulator/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="competency_gap_intelligence">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 24</div>
<h3>Competency Gap Intelligence</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/competency_gap_intelligence/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/competency_gap_intelligence/review_overview.svg?v=20260926-evidence-review" alt="Which evidenced skill gaps should be addressed first? Supplied role targets, dated competency evidence, prerequisites and resource metadata."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/competency_gap_intelligence/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/competency_gap_intelligence/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Observed level = sum(confidence×level)/sum(confidence); priority = positive gap×importance."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Competency Gap Intelligence compares role requirements with dated evidence that carries an explicit confidence label, while keeping missing or stale evidence visible. Only gaps with sufficient support are ranked, after which prerequisite relationships shape development sequences and resource matches. The calculations provide auditable decision support and do not convert subjective competency ratings into validated measurement.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/competency_gap_intelligence/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/competency_gap_intelligence" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/competency_gap_intelligence/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="workplace_learning_recommender">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 25</div>
<h3>Workplace Learning Recommender</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/workplace_learning_recommender/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/workplace_learning_recommender/review_overview.svg?v=20260926-evidence-review" alt="Which feasible resource best addresses a confirmed gap? Supplied competency needs, work context and resource metadata."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/workplace_learning_recommender/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/workplace_learning_recommender/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Resource score = sum(normalized weight × component score)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Workplace Learning Recommender first filters resources for feasibility and then ranks them against confirmed competency gaps and work context. The score exposes contributions from need priority, gap coverage, task fit, quality, effort and other declared factors, followed by coverage and weight sensitivity checks. The output is a recommendation based on supplied metadata, not an estimate of learning impact.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/workplace_learning_recommender/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/workplace_learning_recommender" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/workplace_learning_recommender/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="training_transfer_analytics">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 26</div>
<h3>Training Transfer Analytics</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/training_transfer_analytics/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/training_transfer_analytics/review_overview.svg?v=20260926-evidence-review" alt="Does observed workplace application persist after training? Synthetic baseline and follow-up trajectories with self, manager, behavior and objective evidence."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/training_transfer_analytics/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/training_transfer_analytics/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Persistence ratio = final follow-up application / first follow-up application."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Training Transfer Analytics separates workplace application from transfer conditions and business outcomes. Multiple evidence sources are combined with explicit weights, baseline and follow up changes are tracked, and disagreement among indicators is reported rather than hidden in one composite score. The bundled trajectories are synthetic, so observed persistence is described without claiming that training caused the change.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/training_transfer_analytics/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/training_transfer_analytics" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/training_transfer_analytics/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="engagement_early_warning">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 27</div>
<h3>Engagement Early Warning System</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/engagement_early_warning/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/engagement_early_warning/review_overview.svg?v=20260926-evidence-review" alt="What support signal was available at the prediction date? Synthetic dated activity, task and assessment histories with future outcome windows."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/engagement_early_warning/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/engagement_early_warning/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Risk = 1/(1+exp(-z)); z = intercept + sum(coefficient×available feature)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Engagement Early Warning System enforces a strict time boundary between predictors available at the point of assessment and outcomes observed later. Synthetic risk coefficients are exposed, and a capacity limited review queue is evaluated together with calibration and subgroup diagnostics. The code demonstrates leakage controls and support oriented reporting, but it does not claim predictive validity for real learners.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/engagement_early_warning/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/engagement_early_warning" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/engagement_early_warning/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="curriculum_knowledge_graph">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 28</div>
<h3>Curriculum Knowledge Graph</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/curriculum_knowledge_graph/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/curriculum_knowledge_graph/review_overview.svg?v=20260926-evidence-review" alt="Where are curriculum prerequisites and assessment gaps? Supplied typed entities and provenance-labeled relationships."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/curriculum_knowledge_graph/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/curriculum_knowledge_graph/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Coverage = eligible entities with a required relation / all eligible entities."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Curriculum Knowledge Graph represents courses, concepts, outcomes, assessments and resources through typed relationships with recorded provenance. It checks relation semantics and identifies prerequisite paths, cycles, disconnected entities and gaps in assessment coverage. These structural diagnostics support expert review of a curriculum; the graph does not independently establish the pedagogical validity of its relationships.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/curriculum_knowledge_graph/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/curriculum_knowledge_graph" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/curriculum_knowledge_graph/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="self_regulated_learning_copilot">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 29</div>
<h3>Self Regulated Learning Copilot</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/self_regulated_learning_copilot/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/self_regulated_learning_copilot/review_overview.svg?v=20260926-evidence-review" alt="When should support be offered, delayed or declined? Supplied learning state, explicit plans, preferences, reflection and prompt history."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/self_regulated_learning_copilot/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/self_regulated_learning_copilot/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Plan completeness = present planning fields / 4."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Self Regulated Learning Copilot uses supplied plans, progress, preferences and interaction history to decide whether assistance should be offered or withheld. Refusals and explicit help requests are preserved, prompt burden is limited and reflection is connected to later strategy choices. The examples are synthetic, while plan completeness and struggle labels are operational rules rather than validated psychological measures.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/self_regulated_learning_copilot/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/self_regulated_learning_copilot" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/self_regulated_learning_copilot/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
<article class="repo-project" id="feedback_quality_evaluator">
<header class="repo-project-heading"><div class="repo-kicker">AI in Education Repository · 30 · Final entry reviewed</div>
<h3>Feedback Quality Evaluator</h3></header>
<div class="repo-project-grid">
<div class="repo-visuals">
<figure class="repo-visual"><a class="figure-link" href="assets/aied/feedback_quality_evaluator/review_overview.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/feedback_quality_evaluator/review_overview.svg?v=20260926-evidence-review" alt="Is feedback grounded and actionable in its supplied context? Feedback text, goal, learner work, evidence and an optional external reference verdict."></a><figcaption>Study question, data, design and interpretation limits</figcaption></figure>
<figure class="repo-visual"><a class="figure-link" href="assets/aied/feedback_quality_evaluator/review_calculations.svg?v=20260926-evidence-review" target="_blank" rel="noopener noreferrer" aria-label="Open figure at full size in a new tab"><img loading="lazy" src="assets/aied/feedback_quality_evaluator/review_calculations.svg?v=20260926-evidence-review" alt="Calculation: Kappa = (observed agreement - chance agreement)/(1 - chance agreement)."></a><figcaption>Calculation definition and source linked evidence; open for full size</figcaption></figure>
</div>
<div>
<div class="repo-copy"><p>Feedback Quality Evaluator assesses feedback against supplied task context using separate rubric dimensions. Missing evidence remains unevaluable, lexical triggers are exposed and rater agreement tools handle degenerate cases explicitly. The implementation is a baseline for validation and error analysis rather than an automated authority on disciplinary accuracy or feedback quality.</p></div>
<div class="repo-links"><a href="https://github.com/devissaputra/feedback_quality_evaluator/blob/main/CALCULATIONS.md" target="_blank" rel="noopener noreferrer">Calculations & evidence</a><a href="https://github.com/devissaputra/feedback_quality_evaluator" target="_blank" rel="noopener noreferrer"><i class="fa-brands fa-github"></i> Repository</a>
<a href="https://github.com/devissaputra/feedback_quality_evaluator/blob/main/README.md" target="_blank" rel="noopener noreferrer"><i class="fa-solid fa-file-lines"></i> README</a></div>
</div>
</div>
</article>
</section>
<section id="contact">
<div class="eyebrow-rule"><h2>Contact</h2></div><p>Open to learning design, L&D and learning technology roles, as well as doctoral research in Artificial Intelligence in Education, learning analytics and educational systems centered on human judgment and learning needs. Based in Medan, Indonesia and open to remote work or relocation.</p><div class="contact-row"><a href="mailto:devis.saputra@gmail.com">Email</a><a href="https://www.linkedin.com/in/devissaputra/">LinkedIn</a><a href="https://orcid.org/0000-0002-7133-9410">ORCID</a></div>
</section>
<footer><a href="https://github.com/devissaputra/devissaputra.github.io/blob/main/REPOSITORY_REVIEW.md">Repository review: 39/39 completed</a><br>© Devis Saputra · AI in Education portfolio · Built for GitHub Pages</footer>
</main>
</div>
</body>
</html>