RAG Pipeline
Retrieve trusted context, rank sources, draft grounded answers, and show citations with confidence.
What you build
Four AI engineering systems in one repo.
Each module is small enough to understand and useful enough to extend into a real product.
Prompt Evals
Score model outputs against expected keywords, forbidden claims, and behavior checks.
Safety Scanner
Detect PII, financial, legal, medical, and self-harm risk before responses reach users.
Agent Planner
Turn goals into reviewable workflows with retrieval, drafting, evaluation, and human handoff.
Learning path
A curriculum that ships artifacts.
This is not passive reading. Every phase ends with code, prompts, evals, datasets, or workflow specs that contributors can improve.
Phase 00
Foundations
Understand the GenAI loop: user goal, prompt, context, output, evaluation, safety, and product action.
Runnable proof
One command shows the full lab.
The repo runs without API keys, so anyone can clone it, test it, and contribute immediately.
npm run check
build passed
demo generated
Smoke tests passed.Open source
Built for people who want to learn by shipping.
Add an embeddings lesson, improve retrieval ranking, create eval cases, or build the first browser UI. The repo has starter issues ready.