hipporag

I put all my exploration around AI in reproducible notebooks in this repository

GraphRAG over aviation safety incident reports for causal reasoning and safety pattern extraction

Advance RAG is an enterprise-grade RAG system built on HippoRAG, designed for high-accuracy question answering in multilingual environments. It combines knowledge graph-based retrieval, hybrid search strategies, and advanced query processing to deliver precise, source-cited responses.

One install, three MCPs: graph chat memory + semantic code search + backend. Benchmarked with claude-haiku-4-5: -19% tokens & 23× faster on memory-recall questions.

Autonomous web-pentest agent on consumer hardware — LangGraph + 23-tool MCP server + HippoRAG knowledge graph, driving a fine-tune-free Qwen2.5-Coder-14B on 12GB VRAM. Clears every Low/Medium module and 84.6% at High on DVWA.

Local MCP memory server for LLMs. Bi-temporal graph + vector + Cypher queries + auto-write/auto-recall hooks for Claude Code. Works with Ollama, LM Studio, Anthropic, OpenAI.

Local-only, graph-augmented, iteratively-reasoning, self-corrective RAG showcase on Apple Silicon — pushing toward published SOTA on multi-hop QA (MuSiQue, 2WikiMultiHopQA, HotpotQA) with HippoRAG 2 + Adaptive routing + Search-o1 iterative reasoning + CRAG-style gating + NLI faithfulness verification.

Benchmark comparing Naive RAG, Light RAG, and HippoRAG2-Hier on Japanese river & sediment-control technical standards. 6 conditions (3 RAG × 2 LLMs), AI-as-Judge scoring.

HippoRAG 2 (graph) vs. ChromaDB (vector) retrieval, benchmarked head-to-head on 3 multi-hop QA datasets — real recall, significance tests, and cost

MedGraphia - a Medical Q&A system via GraphRAG and LLMs

A dual-store memory engine that finds what you know, not what you wrote. Semantic search meets knowledge graph — so the right idea surfaces at the right moment.