Software Engineering student · Applied AI & data-intensive systems
Tunis, Tunisia
Final-year Software Engineering student at INSAT (University of Carthage), focused on applied AI and data-intensive systems. Through internships in medical imaging, regulatory document analysis and big-data pipelines, I've learned that I enjoy the research side of engineering most: designing architectures, testing hypotheses, and iterating when the results fall short.
- 🎓 Studying: National Engineering Degree in Software Engineering, INSAT (2022–2027)
- 🧠 Interests: NLP and LLMs, medical imaging, graph-based reasoning, big data and streaming
- 🏁 Outside class: CTF competitions (web exploitation) and competitive programming
- 🗣️ Languages: Arabic (native), English (advanced), French (intermediate), German (beginner)
| Project | What it does | Highlight |
|---|---|---|
| document-incoherence-detection | Finds contradictions between QHSE documents. A Neo4j dependency graph selects which clauses to compare, then rules → NLI → LLM judge verify them with verbatim evidence | 95 % of pairs pruned, 19/19 inconsistencies kept, 16/19 found (F1 0.80) on a small synthetic corpus |
| dental-xray-anomaly-detection | Two-stage pipeline on panoramic X-rays: U-Net + ResNet-34 tooth segmentation, then EfficientNet-B3 per-tooth diagnosis | Dice 0.84, handles a ~128:1 class imbalance; evaluated on 50 unseen X-rays |
| fer2013-emotion-recognition-benchmark | Six models compared on FER2013, plus a Streamlit app to test them | A 5.87M-parameter custom CNN reaches 66.9 % accuracy (0.651 macro-F1), beating VGG16 and ResNet50 |
| qhse-document-rag | French question answering over QHSE documents: FAISS retrieval + Google Gemini, answers grounded in and cited from the sources | Strict prompt: context-only answers, mandatory citations, explicit "I don't know" |
| music-analytics-lambda-pipeline | Lambda-architecture pipeline on 550K Spotify tracks: Kafka, HDFS, Spark batch + streaming, Cassandra / PostgreSQL / Elasticsearch, Grafana | Real-time mood tracking, trend and anomaly detection, K-Means genre clustering |
| film-revenue-prediction | Six regressors benchmarked to predict box-office revenue from pre-release data, with a Flask app | Random Forest R² 0.751; engineered a Bayesian cast star-power feature; Gaussian Process uncertainty |
Also built: DoussiyaFidaya, a full-stack medical platform (NestJS, Next.js, PostgreSQL) with role-based spaces, 2FA and an AI layer (OCR + RAG chatbot), and a package delivery optimiser implementing six search algorithms from scratch in Java.
R&D Software Engineering Intern, Saphir Consult (Qualipro), Tunis
- Jul–Aug 2026: architected a cross-document contradiction detector for QHSE normative documents (Neo4j dependency graph, spaCy → NLI → LLM cascade).
- Jul 2025: built a RAG question-answering system (Sentence Transformers, FAISS, Gemini API) with a Flask web app.
Languages: Python · JavaScript · TypeScript · Java · C++ · C# · C · PHP · SQL
Also: Sentence Transformers · RAG · LLaMA 4 · Mistral 7B · Gemini API · LM Studio · OCR · MLlib · HDFS · MySQL · SQL Server · Metabase · pytest / TDD
- HackerRank: Software Engineer Intern (Java, Python, REST API)
- Google: Cloud Computing Foundations (Cloud Computing Fundamentals, Infrastructure in Google Cloud)
Let's connect: I'm happy to talk about NLP, applied AI and data systems.
LinkedIn ·
Email