Imported from davidcp90/raglab-ai4devs (AGENTS.md). Install upstream with npx skills add davidcp90/raglab-ai4devs. Copyright stays with the author.

Agent guidance (AGENTS.md)

Context for AI agents working in this repo: layout, key files, how to run things, and where to change behavior.

Project overview

• Goal: ReAct chatbot (Wikipedia + RAG) served by Flask, with embeddings stored in PostgreSQL (pgvector).
• Source of truth: practica-raglab-ai4devs.ipynb — Paso 3 (RAG/embeddings) and Paso 4 (ReAct + Flask).
• Secrets: .env (gitignored); template is .env.example.

Layout

raglab/
├── .env.example          # Env template (OPENAI_API_KEY, POSTGRES_*)
├── .env                  # Real secrets (do not commit)
├── docker-compose.yml    # postgres (pgvector) + flask
├── Dockerfile            # Flask app image
├── requirements.txt     # Python deps
├── app/
│   ├── __init__.py
│   ├── app.py            # Flask app, PGVector retriever, qa_chain, ReAct agent
│   ├── config.py         # Env loading, DATABASE_URL, COLLECTION_NAME
│   └── templates/
│       └── index.html    # Chat UI
├── static/
│   └── styles/
│       └── name.css      # Optional overrides for UI
├── scripts/
│   └── upload_embeddings.py   # Paso 3 → PGVector (docs, split, embed, store)
├── index.html            # Reference copy (UI lives in app/templates/)
└── practica-raglab-ai4devs.ipynb   # Original notebook

Key files

How to run

• Start stack: docker compose up -d
• Upload embeddings: docker compose run --rm flask python scripts/upload_embeddings.py
• Local Flask (with Postgres on host): set DATABASE_URL or POSTGRES_HOST=localhost in .env, then python -m app.app
• Local upload: same env; python scripts/upload_embeddings.py

Conventions

• Use libs already in the repo (see requirements.txt). Prefer langchain_community PGVector with psycopg2-binary unless switching to langchain-postgres/psycopg3 is required.
• Config and secrets: env only (.env); no hardcoded API keys or DB URLs.
• Vector collection name is shared: raglab_docs (or PGVECTOR_COLLECTION_NAME). Upload script and Flask app must use the same name.

Where to change what

Architecture (reminder)

1. Upload: upload_embeddings.py → same docs as Paso 3 → split → OpenAI embeddings → PGVector (raglab_docs).
2. Runtime: Flask starts → PGVector store (same collection) → retriever → qa_chain (RAG) → Wikipedia + RAG tools → ReAct agent; /ask_bot runs agent.run(msg).

Keep README and this file in sync when adding features or changing layout.