"I don't think I've typed a line of code probably since December." β Andrej Karpathy
The future PI doesn't run experiments, doesn't write first drafts, doesn't draw figures. They make decisions. The AI agent team handles the rest.
Academic GStack applies Garry Tan's gstack philosophy to scientific research: gstack turns one engineer into an engineering team. Academic GStack turns one researcher into a virtual lab.
Every scholar = an academic personal brand startup.
- Your paper = your product
- Your citations = your user growth
- Your h-index = your valuation
- Your lab = your company
π Get Started in 5 Minutes β | Tutorial | Examples | FAQ
| gstack (Software Engineering) | Academic GStack (Scientific Research) |
|---|---|
/office-hours β product direction interrogation |
/lab-meeting β research direction interrogation |
/plan-ceo-review β CEO scope challenge |
/pi-review β PI strategic review |
/plan-eng-review β architecture lock-in |
/method-review β experimental protocol lock-in |
/plan-design-review β design review |
/figure-review β figure quality review |
/review β code review |
/peer-review β simulated peer review |
/qa β browser testing |
/data-qa β data quality audit |
/cso β security audit |
/novelty-check β novelty verification |
/ship β release engineer |
/submit β submission engineer |
/retro β engineering retro |
/lab-retro β project retrospective |
/codex β second opinion |
/second-opinion β independent AI peer review |
/autoplan β auto-review pipeline |
/auto-experiment β auto-experiment pipeline |
| Role | Function |
|---|---|
| PI Reviewer | Scope challenge: is this direction worth pursuing? What tier journal? |
| Lab Strategist | Multi-project scheduling: resource allocation, priorities, deadline management |
| Grant Scout | Match funding opportunities, auto-generate proposal drafts |
| Role | Function |
|---|---|
| Lit Reviewer | Systematic search + auto-classification + extract testable hypotheses |
| Hypothesis Generator | Generate ranked hypotheses from known data |
| Method Designer | Design experimental protocols, estimate sample sizes, write protocols |
| Data Analyst | Auto statistics, visualization, anomaly detection, effect size calculation |
| Data QA | Audit data integrity, find flaws, verify reproducibility |
| Figure Artist | Auto-generate publication-quality figures |
| Novelty Checker | Web-wide novelty search, flag existing work |
| Role | Function |
|---|---|
| Paper Drafter | Write first draft: AbstractβIntroβResultsβDiscussion |
| Peer Reviewer | Simulate 3 reviewers, provide revision feedback |
| Journal Matcher | Analyze paper features β recommend target journals + acceptance estimate |
| Grant Writer | Auto-generate grant proposals from papers |
| Cover Letter | Auto-write cover letters for target journals |
| Rebuttal Drafter | Analyze reviewer comments β generate response strategy |
| Role | Function |
|---|---|
| Lab Manager | Multi-project status board, deadline alerts, resource conflict detection |
| Lit Monitor | arXiv/bioRxiv new paper alerts + relevance scoring |
| Social Writer | Auto-generate Twitter/X threads + LinkedIn + popular science summaries |
| Lab Archivist | Auto-organize lab notes, code, data into structured knowledge base |
Think βββ Plan βββ Experiment βββ Analyze βββ Write βββ Review βββ Submit βββ Reflect
β β β β β β β β
/lab- /method- Auto-Exp /data- /paper- /peer- /submit /lab-
meeting review analysis draft review +cover retro
β β β β β β β
/pi- /method- /data-qa /figure- /peer- /journal- /social-
review review audit artist review match writer
β β β
You decide CHECKPOINT You decide CHECKPOINT
- Design doc complete (18 role definitions + full workflow)
- P0 skills complete (Lit Reviewer / Hypothesis Generator / Peer Reviewer)
- Strategy layer complete (PI Reviewer / Lab Strategist)
- Pipeline engine complete (ac-gstack-pipeline + cron)
- End-to-end deployment (Hermes Skills + nightly cron jobs)
- All 18 skills complete (100%)
- 3 cron jobs active (nightly pipeline + lit monitor + grant scout)
- Documentation (Getting Started + Tutorial + Examples + FAQ)
- Video tutorial / docs site (future)
18/18 skills complete (100%) π
- Agent Framework: Hermes Agent
- Science Engine: Co-Scientist (hypothesis generation + debate)
- Orchestration: maestro (multi-agent task dispatch)
- Scheduling: cron + delegate_task
- Garry Tan's gstack β the one-person engineering team
- Andrej Karpathy β four rules of AI-assisted programming
- Science CEO Methodology β from lab technician to research commander
MIT License Β· Building ποΈ
git clone https://github.com/SpencerRaw/academic-gstack.git
cp -r skills/* ~/.hermes/skills/Requires Hermes Agent. See Getting Started for full setup.
hermes skill load lit-reviewer
hermes skill load peer-reviewer
hermes skill load hypothesis-generator
See CONTRIBUTING.md. New skills follow the skill template.