← Back to Building
Personal

Knowledge-Graph Reasoning Agent

Hybrid graph and vector reasoning.

Overview

A compact solo build examining when symbolic graph traversal beats neural retrieval, and how to combine them. An agent decides, per query, whether to traverse a structured knowledge graph, search a vector store, or do both.

Approach

A multi-model database holding documents, graph, and vectors together; a small set of graph-reasoning tools (multi-hop traversal, centrality, path-finding, community detection); and a planner that selects the right mode per query. Entirely local.

Scope

Small hobby project, built solo in a few days.