ContRAG

Overview

RAG engine that introspects your database schema at runtime, builds an entity relationship graph, and powers Dhaniverse's adaptive AI NPCs.

Highlights

• Introspects an existing database schema at runtime and builds an entity relationship graph, so retrieval follows real relationships instead of similarity alone.
• Tracks per-user preferences across turns for personalized context retrieval.
• Works across MongoDB, PostgreSQL, and vector stores, with a Rust core published to crates.io.

Reading

The entity-graph retrieval idea sits on top of a body of research I keep coming back to:

• Retrieval-Augmented Generation: A Comprehensive Survey of Advances and Applications in Deep Learning Models
• A Survey of Agentic GraphRAG: From Retrieval-augmented Generation to Graph-native Agents
• High-Dimensional Vector Similarity Search: From Time Series to Deep Network Embeddings (SIGMOD 2020)

Tech stack