Revolutionizing RAG for Financial Data With Knowledge Graphs
summary
Demo shows an end-to-end pipeline that ingests PDFs, slides and XBRL, auto-builds an ontology, and stores facts in Neo4j while using Azure OpenAI embeddings and a vector store for retrieval. Claims a jump from a baseline 53–60% acceptance with a pure vector RAG to ~96% acceptable/correct answers with a hybrid Neo4j graph RAG, typically a 35%+ accuracy uplift. Live ChatQA ties answers back to graph nodes for traceability and auditability.
AI-written summary · about
chapters (8)
- 0:00Introduction to Lettria
- 1:50Challenges in document parsing
- 4:22Limitations of vector RAG
- 7:14GraphRAG performance benefits
- 10:07How GraphRAG works
- 14:05Live demo and benchmarking
- 17:56Methodology and deployment
- 20:11Company overview and closing
related talks
39mEnhancing RAG with Neo4j Knowledge Graph
30mAgentic RAG with Knowledge Graphs Explained
28mData-Empowered AI Assistants
25mAdaptive GraphRAG: A Framework for Knowledge Graph Quality, Consistency, Evolution
12mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
12mRevolutionizing Platform Engineering with AI Agents and GraphRAG
21mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval
27mGraphReader Agentic RAG: Rethinking Long-Context Retrieval Systems