TraceGraph: Graph-Based Multi-Hop Reasoning for RAG

This presentation introduces TraceGraph, a graph-based retrieval system designed to improve multi-hop reasoning in Retrieval-Augmented Generation (RAG). By modeling documents as interconnected graphs and enabling iterative retrieval through a revisitable agent, TraceGraph produces more accurate, explainable, and auditable answers.

We demonstrate the system architecture, core AI techniques, evaluation results, and a working prototype that highlights improved evidence coverage and reasoning capabilities over traditional RAG approaches.

Course: CECS 551 – Artificial Intelligence
Team: Utkarsh Balu Lubal · Kashif Manzer · Mayank Maurya · Dipak Yadav

Видео TraceGraph: Graph-Based Multi-Hop Reasoning for RAG канала Utkarsh Lubal