Projects

GraphRAG Research Assistant

Multi-hop RAG system with hybrid graph + vector retrieval using an LLM knowledge graph and pgvector. It doubled answer accuracy from 19% to 38%, raised recall@5 to 78%, and cut retrieval latency from 6.6s to 2.3s.

View project

Relay

Multi-provider LLM gateway with semantic caching, ONNX embeddings, Redis-stack vector search, rate limits, budgets, cost/latency routing, automatic failover, SSE streaming, and cost analytics.

View code