TechRAG: Evidence-Gated Multimodal Agentic RAG for Technical Literature Reasoning

The paper introduces TechRAG, an evidence-gated multimodal retrieval-augmented generation (RAG) framework designed for reasoning over technical literature in fields like intelligent tires and vehicle dynamics. It features a sophisticated architecture incorporating hybrid text retrieval methods (FAISS and BM25), a Neo4j knowledge graph for evidence expansion, and a multi-agent system that includes self-correcting revision capabilities. This framework enhances the retrieval process by integrating text, visual data, and graph evidence, making it a significant advancement for practitioners aiming to improve the accuracy and depth of literature reasoning in specialized domains.