Skip to content

Adaptive RAG with self-reflection and external tools

Notifications You must be signed in to change notification settings

stanislavlia/deepRAG

Repository files navigation

How to run

docker compose -f docker-compose-stage.yaml up

Web UI will be located at localhost:8051

Usage

example

TODOs

current tasks:

  1. Write unit-tests for backend and put them to CI pipeline
  2. Migrate from Streamlit to classic frontend (NextJS + HTML)

Long-term imortant things:

  1. Switch from streamlit to nextJS and normal frontend
  2. Add LLM and turn app to be RAG (Retrieval Augmented Generation)
  3. Store files in localstack (local AWS S3)
  4. Add authorization
  5. Add other formats apart PDF (doc, txt, pptx etc)
  6. Migrate fully to Langchain - DONE
may be needed later:
  1. Store logs in ELK stack
  2. Handle multiple collections
  3. Record info about users to Relational DB

Notes for improving RAG

  1. Query augmentation with LLM
  2. Cross encoder to score retrieved docs
  3. Use langsmith to monitor chain
  4. Routing between databases/promts