![](data:image/svg+xml;utf8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none" stroke="%23B8593E" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round"><circle cx="12" cy="12" r="3"/><path d="M12 2v3M12 19v3M2 12h3M19 12h3"/><path d="M5.6 5.6l2.1 2.1M16.3 16.3l2.1 2.1M5.6 18.4l2.1-2.1M16.3 7.7l2.1-2.1"/></svg>)
Welcome to Learn AutoResearch
Learn AutoResearch is a project-based course on automating research using the autoresearch framework — a generalization of Karpathy's autonomous ML training loop to any domain with a measurable metric.
"Set the GOAL → The agent runs the LOOP → You wake up to results."
What you will learn
- Define measurable research goals — turn vague objectives into mechanical metrics any agent can optimize.
- Run autonomous improvement loops — one change per iteration, automatic rollback, git as memory.
- Debug scientifically — falsifiable hypotheses, evidence-based investigation, zero-error termination.
- Predict before acting — five expert perspectives before committing to any major change.
- Audit security autonomously — STRIDE + OWASP + red-team analysis with code-level evidence.
- Ship with confidence — 8-phase pipeline covering code, content, and deployments.
Get started
Lectures
12 lectures from first principles (why manual iteration fails) to advanced overnight runs and CI/CD integration.
Projects
Six hands-on projects — each with a starter codebase and a reference solution, building to a full end-to-end pipeline.
Resource Library
Copy-ready templates: research.md, evaluate.py, results.tsv, and metric cheat sheets for 15 domains.
The Core Loop
Every autoresearch command is built on the same five-stage loop:
Course Structure
The course is organized into 6 phases, each containing 2 lectures and 1 hands-on project:
| Phase | Theme | Lectures | Project |
|---|---|---|---|
| 1 | Why AutoResearch Works | L01–L02 | Sort optimization |
| 2 | Master the Core Loop | L03–L04 | Function fitting |
| 3 | Debug & Fix | L05–L06 | FastAPI debugging |
| 4 | Predict & Reason | L07–L08 | Architecture debate |
| 5 | Security & Scenarios | L09–L10 | Security audit pipeline |
| 6 | Ship & Advanced Patterns | L11–L12 | End-to-end research |
Next steps
- Lecture 01: Why Manual Iteration Fails — Start with Karpathy's original insight.
- Project 01: Your First Research Loop — Run the sort optimization example hands-on.
- Templates — Grab the research.md and evaluate.py templates for your own projects.