ai-learning

Local-first playground for learning LLM apps: Ollama, prompting, tools, RAG, embeddings.

Layout

ai-learning/
├── docker-compose.yml     # Ollama → ./models
├── .env.example           # copy to .env
├── models/                # Ollama data (volume)
├── apps/
│   ├── chat-cli/          # thin Node client — start here
│   └── api-server/        # add later for HTTP/UI
├── experiments/           # numbered steps (01 → 05)
├── notebooks/             # optional Jupyter
├── data/raw|processed/    # datasets (heavy stuff gitignored)
├── prompts/               # versioned system / few-shot text
├── docs/                  # internal runbooks & notes (not the public site)
└── apps/docs/             # Docusaurus site (roadmap, journal, AI notes)

Learning path (order)

Step	Folder	Focus
1	experiments/01-prompting	Ollama chat, params, logging
2	02-tools-function-calling	tool schemas, parse, round-trip
3	03-rag-chunking	chunk, embed, retrieve, cite
4	04-embeddings-similarity	similarity, failure modes
5	05-fine-tuning-or-lora	optional, only if needed

Implement scripts in each experiments/NN-* folder; reuse patterns from apps/chat-cli when it makes sense.

Quick start

cp .env.example .env
docker compose up -d
ollama pull llama3.2   # or pull inside container: docker exec -it <ollama> ollama pull llama3.2
cd apps/chat-cli && node index.js

If Ollama runs on the host instead of Docker, point OLLAMA_HOST in .env accordingly.