[Perf] #93 — Cache document embeddings in Redis to prevent redundant API/encoding calls

[adarsh-gautam-sys]

Summary

The RAG pipeline generates document embeddings repeatedly for identical documents and text queries, resulting in redundant encoding work and increased API/computational costs. This PR introduces a Redis caching layer for embeddings, storing the generated vector embeddings with a 24-hour expiration (TTL).

Changes Made

• app/rag/embeddings.py:
  • Added a synchronous Redis client with connection pooling and a 2s connection timeout.
  • Implemented md5 text-hashing to generate unique cache keys (embedding:{text_hash}).
  • Added lookup and set operations on Redis before/after encoding.
  • Handled exceptions gracefully to fallback to standard encoding in case of connection failure or timeouts.

Related Issue

Closes #93

[vercel]

@adarsh-gautam-sys is attempting to deploy a commit to the devloperdevesh's projects Team on Vercel.

A member of the Team first needs to authorize it.