GraphMind
AI-powered knowledge system
Transforms unstructured documents into connected, queryable knowledge using a Hybrid Retrieval (Graph + Vector) reasoning system.
Problem
Standard RAG (Retrieval-Augmented Generation) fails on complex reasoning tasks because it lacks understanding of global relationships between concepts.
Approach
Conceptualized a solution to the 'multi-hop' reasoning problem by explicitly defining relationships in a graph database, combined with pgvector for Hybrid Retrieval.
What shipped
- Successfully extracted strict semantic triplets (Subject-Predicate-Object) from messy text.
- Proved that Hybrid retrieval (Graph + Vector) dramatically outperforms pure vector search for multi-hop synthesis.
Stack
Next.js
FastAPI
Python
Postgres
Neo4j