Constitución Colombia Chat

A Streamlit RAG chat over the Constitución Política de Colombia de 1991.

The app indexes the constitution by article, stores the embeddings in a vector database, and answers questions using only the retrieved articles.

Quickstart

cp .env.example .env
uv sync
uv run python scripts/vectorize_constitution.py
uv run streamlit run main.py

Open the app at the URL printed by Streamlit, usually:

http://localhost:8501

Docker Compose quickstart

cp .env.example .env
# Edit .env with your OpenAI-compatible endpoint and Qdrant settings.
docker compose build
docker compose up

Open the app at:

http://localhost:8501

The Docker entrypoint runs scripts/vectorize_constitution.py before starting Streamlit. Ingest is idempotent: it creates the collection if missing, recreates it only if the embedding dimension changes, and upserts all articles otherwise.

Docker/Qdrant settings

For a remote Qdrant instance:

VECTOR_DB_PROVIDER=qdrant
VECTOR_DB_URL=https://qdrantinstance.com/
VECTOR_DB_PORT=443
VECTOR_DB_API_KEY=...
VECTOR_DB_COLLECTION_NAME=colombia_constitution

For the included local Qdrant service:

VECTOR_DB_URL=http://localhost:6333
VECTOR_DB_PORT=
VECTOR_DB_API_KEY=

The compose file maps:

• Streamlit: http://localhost:8501
• Qdrant REST UI/API: http://localhost:6333

Do not commit .env; it contains API keys.

Requirements

• Python 3.14
• uv
• Docker Compose, for containerized runs
• A running vector database (default: Qdrant at http://localhost:6333)
• An OpenAI-compatible API endpoint that supports both embeddings and chat

Configuration

Edit .env:

OPENAI_URL=...
OPENAI_KEY=...

VECTOR_DB_PROVIDER=qdrant
VECTOR_DB_URL=http://localhost:6333
VECTOR_DB_PORT=
VECTOR_DB_API_KEY=
VECTOR_DB_COLLECTION_NAME=colombia_constitution

EMBEDDING_MODEL=...
CHAT_MODEL=...
RETRIEVER_TOP_K=5

How it works

1. scripts/vectorize_constitution.py extracts the PDF text.
2. The text is split into one chunk per article.
3. Each chunk is embedded and upserted into the vector database.
4. main.py streams the Streamlit chat.
5. Each question retrieves relevant articles and passes them to the chat model.

Development

• uv sync — install dependencies
• uv run ruff check . — lint
• uv run ruff format . — format
• codegraph sync — refresh local CodeGraph index

See AGENTS.md for repo-specific instructions for future coding sessions.