multi-hop-qa

7-layer RAG framework that eliminates semantic drift + context poisoning. Faithfulness 0.94 vs 0.71 Naive RAG. +13.6 EM on multi-hop QA.

Evaluation code for studying how upstream ASR errors affect multi-hop retrieval-augmented generation.

Knowledge-graph RAG library for multi-hop QA — iterative retrieve-and-reason (IRCoT), graph retrieval, and cross-document enrichment at ingest. Production retrieval pipeline in Python.

[EMNLP 2026] Official implementation of SAFE: An LLM-as-Verifier Framework for Evidence-Grounded Multi-Hop Reasoning.

Binary evidence-sufficiency dissociation in reasoning-model hidden states for fixed-question, changed-context multi-hop QA.

Knowledge-graph RAG for multi-hop QA: a controlled study that overturned its own +8.8 win under realistic evaluation.

A per-instance cost model for AI: compute and verification scale with the difficulty of one question.

A fine-tuned Qwen2.5-1.5B model that breaks complex multi-hop questions into atomic sub-queries with dependency graphs, doubling RAG retrieval accuracy in ~50ms

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

Evidence-based arbitration for multi-hop RAG/QA using FLARE, IRCoT, and FARR candidate trajectories.

Graph-based RAG: documents -> LLM-extracted knowledge graph -> multi-hop, cited answers with an interactive network view. LangChain + NetworkX + Streamlit.

Multi-agent system for multi-hop information retrieval: query decomposition, tool-augmented retrieval, knowledge-graph traversal, and self-reflective validation. IR course project, Università di Napoli Federico II.

Selective local graph rescue after hybrid retrieval for multi-hop RAG, with calibrated gating and equal-budget evaluation.

Official implementation of Cascading Hallucination in Agentic RAG: The CHARM Framework for Detection and Mitigation.

Clean research implementation of Bactrainus for modular multi-hop question answering.

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.

Per-chunk RAG verification fails on multi-hop questions. Entailment: 0.951 AUC single-hop, 0.643 multi-hop. Full traces released.