Generative AI Engineering Lab

Learn Generative AI engineering by building the systems behind real products.

Quick Start . What You Build . Curriculum . Good First Issues

Why This Repo Exists

Most people do not need another vague AI list. They need a practical path from idea to working system:

• retrieve trusted context
• generate grounded answers
• evaluate outputs
• scan safety risks
• plan agent workflows
• ship with tests and CI

This repo is small enough to understand, but structured like a serious open-source project.

Quick Start

npm install
npm run check

Run demos one by one:

npm run demo:rag
npm run demo:evals
npm run demo:safety
npm run demo:agent
npm run eval:report
npm run build && node dist/src/eval-report-cli.js --json

Run the website:

npm run site

Then open http://localhost:4173.

What You Build

Module	What It Teaches	Working Code
RAG	Retrieval before generation, citations, confidence	src/lab.ts
Prompt Evals	Expected behavior, forbidden claims, scoring	src/lab.ts
Safety Scanner	PII, legal, medical, financial, self-harm risk	src/lab.ts
Agent Planner	Multi-step tool workflows and human handoff	src/lab.ts
Production Basics	CI, tests, docs, issue templates, roadmap	.github/, tests/

Demo Preview

=== Generative AI Engineering Lab ===
rag: cited answer with confidence
evals: prompt quality test passed
safety: high-impact financial risk detected
agent: five-step workflow generated

Repository Structure

src/                      TypeScript implementation
site/                     Static website
tests/                    Smoke tests
curriculum/               Learning path
prompts/                  Reusable prompt templates
evals/                    Prompt evaluation cases
datasets/                 Sample local knowledge base
outputs/                  Reusable artifacts
docs/                     Roadmap, architecture, contribution ideas
.github/                  CI, issue templates, PR template

Curriculum

The course path is intentionally practical:

1. GenAI foundations
2. Prompt engineering
3. RAG systems
4. Agentic workflows
5. Evals and safety
6. Production patterns

Every phase should produce an artifact: a prompt, eval, dataset, agent spec, or code change.

Roadmap

• Add embedding-based retrieval.
• Add OpenAI/provider adapter while keeping mock/local mode.
• Add more eval cases.
• Add a small browser UI.
• Add MCP server example.
• Add lesson quizzes.
• Add hosted demo.

Current release: v0.1.0

Contributing

Start with docs/good-first-issues.md.

The best first PRs:

• Add one lesson.
• Add one eval case.
• Add one sample dataset.
• Improve RAG ranking.
• Add a small CLI command.

License

MIT. Fork it, learn from it, build on it.