Skip to main content

​
Why Docker?

The RAG pipeline uses BAAI/bge-small-en-v1.5 to embed financial summaries into vectors. This model runs in a lightweight Docker container so you don’t need to install PyTorch on the main server.

​
Start the embedding server

cd backend/docker
docker compose up -d
This starts a FastAPI server on port 8001 that exposes: 
POST /embed
Body: {"texts": ["Daily summary for April 18..."]}
Response: {"embeddings": [[0.12, -0.34, ...], ...]}

​
Verify it’s running

curl -X POST http://localhost:8001/embed \
  -H "Content-Type: application/json" \
  -d '{"texts": ["test"]}'
Should return a 384-dimensional float array.

​
Memory requirements

ModelMemory
BAAI/bge-small-en-v1.5~180 MB RAM
Docker overhead~100 MB
Total~280 MB
A t3.small (2 GB RAM) can comfortably run both the FastAPI backend and the embedding server.

​
Without Docker

If you can’t run Docker, RAG ingestion will fall back to a no-op and the AI advisor will work without memory context. All other features work normally.