World's Fair 2024
Knowledge Graphs & GraphRAG: Techniques for Building Effective GenAI Applications: Zach Blumenthal
Overview
This talk explores building effective Generative AI applications using Knowledge Graphs and Graph Retrieval Augmented Generation (GraphRAG). It demonstrates how to combine vector search with graph traversal and graph embeddings to create more personalized and contextually relevant AI-driven responses. The session focuses on practical implementation using Neo4j, LangChain, and OpenAI, showcasing a fashion recommendation email generation use case.
Who should watch
- AI Engineers
- Product Managers
- Builders working on GenAI applications
- Those interested in enhancing LLM responses with structured data
- Developers looking to implement personalized recommendation systems
Key takeaways
- GraphRAG combines vector search, knowledge graph traversals, and graph embeddings for richer AI applications.
- Neo4j's graph database capabilities are leveraged for storing and querying structured data, enabling sophisticated relationship analysis.
- Vector search, powered by embeddings, finds semantically similar items, while graph traversals incorporate structured data like co-purchase behavior for personalization.
- Graph embeddings, generated using algorithms like Fast RP, can infer new relationships and cluster similar items based on their position and connections within the graph.
- A practical example demonstrates building an AI fashion assistant that generates personalized recommendation emails by integrating vector search, graph traversal, and graph embeddings within an LLM chain.
- The approach allows for dynamic personalization, where recommendations change based on customer ID, purchase history, and even seasonal context.
- LangChain is used to orchestrate the retrieval and generation process, connecting different data sources and LLM calls.
Notable quotes
*Retrieval augmented generation is just when you have this step in the middle when you're communicating with a large language model where instead of sort of like in a chat GPT like way where you're just sending a a question and getting a response back you're going to have that llm application go and pull relevant data through some query.*
*The idea behind this right is again we're combining sort of this knowledge graph which Prov this context and this enrichment with the large language model giving that reasoning and that creativity.*
*Embeddings here basically is helping you kind of scale these other graph traversals right to go because when we do that embedding what we're basically doing is we're taking a graph traversal that's similar to the one we went over and going multiple hops.*
Unofficial community note. Prefer the recording for nuance.