Personal Knowledge Base
Cloud-hosted knowledge base with hybrid retrieval, Google Drive sync, and Voyage AI embeddings for personal document search.
Challenge
I wanted a private, searchable knowledge base backed by my own documents — Google Docs, PDFs, spreadsheets — with retrieval quality beyond basic keyword or single-vector search.
Solution
Built hybrid retrieval pipeline combining dense vector search (pgvector) with PostgreSQL full-text search, fused via Reciprocal Rank Fusion (RRF)
Applied Voyage AI reranking as a final stage to maximize result relevance
Implemented incremental Google Drive sync with folder change detection and category-based document filtering
Exposed an OpenAI-compatible chat completions endpoint with RAG injection
Deployed to GCP Cloud Run with a Neon PostgreSQL database as the vector store
Result
Deployed and in active use. Powers document retrieval for the Sazed agent.