llm-learning

Learn about LLMs and how to interact with them through RAG/MCP/Agents.

This repo isn't the most beautiful engineered code, since it's meant for learning and experimentation. But there are some coding standards to follow, which are outlined in the Python instructions.

Workspace structure

This repo is a uv workspace — a single lockfile (uv.lock) at the root, with independent packages as members.

llm-learning/          ← workspace root (shared lockfile + dev tools)
├── rag/               ← RAG pipeline (Ollama + pgvector)
└── agents/            ← Agent systems (LangGraph)

Dependencies are resolved together and pinned in one place, so all members stay consistent.

Running commands

# Run a command scoped to a specific member (from anywhere in the repo)
uv run --package rag rag --help
uv run --package agents agents --help

# Or cd into the member directory and run directly
cd rag && uv run rag query "Do statins reduce atrial fibrillation?"

# Sync all members at once
uv sync --all-packages

Dev tools

Tools in the root [dependency-groups] (mypy, ruff) apply to the whole workspace:

uv run ruff check .
uv run mypy .