Enhancing Chat-Based Customer Recommendations With GraphRAG and Hybrid Search
summary
GraphRAG uses Neo4j + Qdrant + LLMs to turn chat queries into precise product picks by combining graph-filtering (metadata, schema-flexible specs) with vector similarity and a re-ranking step; pipelines ingest S3/local catalogs into embeddings, then a classifier generates structured cipher queries for hybrid GraphRAG lookup and returns sorted comparisons (links, price, images). Supports four shopping journeys (explore, specific item, review aggregation, on-site actions), calls external shops like Shopee/Momo/Amazon, and relies on multi-stage prompt engineering and a multi-agent action layer.
AI-written summary · about
chapters (5)
- 0:00Introduction and overview
- 3:20Building the recommendation framework
- 8:50Multi-scenario user shopping demos
- 13:29System design and workflow
- 16:46Key takeaways and optimization loop
related talks
01h 40mKnowledge Graphs & GraphRAG: Techniques for Building Effective GenAI Applications
30mThe GraphRAG System That Asks Back: Intelligent Probing with Better Contextual Answers
16mEnhancing Retrieval-Augmented Generation with GraphRAG Patterns in Neo4j
27mAdaptive Knowledge Graphs for Dynamic RAG Systems
35mNatural and conversational shopping experiences
22mAI-Powered Personalized Discovery with Graph Search and VLM-Enhanced Embeddings
19mBeyond Retrieval: Adding a Reasoning Layer to RAG
16mGraphRAG for Supply Chain Resilience: Building an AI Supplier Discovery Engine