Building a HybridRAG Document Question-Answering System
summary
Combining vector RAG retrieval with an explicit knowledge graph produces a hybrid system that raises correct-answer rates in a law/finance benchmark from ~57% to ~81% and pushes combined correct+acceptable to ~90%; implementation uses LLMs to extract triplets, parallel calls to a vector store and a graph DB, and experiments with AWS Bedrock agents and Amazon Neptune for automated graph construction; practical results show higher answer precision but added KG construction effort, domain tuning, retrieval weighting and nontrivial cost.
AI-written summary · about
chapters (14)
- 0:00Introduction to Hybrid RAG
- 1:41Agenda and speaker intro
- 2:31Understanding traditional RAG
- 3:55Limitations of vector RAG
- 5:58What is Graph RAG?
- 7:09Graph RAG construction
- 8:40Graph RAG challenges
- 10:29Combining systems: Hybrid RAG
- 12:32Hybrid RAG use cases
- 14:16Building Hybrid RAG
- 17:05Technical approaches
- 19:27Development options
- 22:49Key takeaways
- 23:51Questions and answers
related talks
20mHybridRAG: A Fusion of Graph and Vector Retrieval
21mBeyond RAG: Building Robust Knowledge Frameworks for Complex Multi-document Retrieval for Enterprise
16mWhen Vectors Break Down: Graph-Based RAG for Dense Enterprise Knowledge
51mDeveloping a Custom RAG System
30mGraph RAG: Bringing together graph and vector search to empower retrieval
25mAdvanced RAG Systems: Scaling Retrieval-Augmented Generation in Production
53mPractical GraphRAG - Making LLMs smarter with Knowledge Graphs
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions