Just Add an LLM: Creating a New Food Recommendation Engine From an Old Cypher Based One
summary
Upgrades a Cypher-only food recommender into three concrete variants: baseline hard-coded Cypher, LLM-generated Cypher, and retrieval-augmented generation with Neo4j Graph RAG. Uses Llama 3 (8B), Python and LangChain to embed food and meal nodes; embeddings produce some surprising neighbors (raw apple linking to turnip). Finds RAG plus constrained Cypher yields accurate, auditable takeout lunches with calorie estimates, while pure LLM outputs hallucinate, don’t match the database, and require careful schema names and example prompts.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:02Graph data modeling
- 3:01Graph analysis approach
- 4:31Traditional recommendation
- 5:45Integrating LLMs
- 8:47Modernizing requirements
- 10:32LLM knowledge exploration
- 12:36Vector-based retrieval
- 14:13Constrained vector search
- 16:03Text-to-Cypher generation
- 19:33Lessons and conclusions
- 21:45Future work and wrap-up
related talks
31mEnhancing RAG-based apps by constructing and leveraging knowledge graphs with open-source LLMs
44mGive Your LLMs a Left Brain
52mGive Your LLMs a Left Brain
36mGraphRAG Explained 🔥 Building Smarter LLMs
28mEnhancing LLMs with Graph Technology
39mSelf-Service Analytics using LLMs
22mBuilding Contextual Knowledge Graphs to Adapt, Evaluate, and Leverage LLMs
30mLLM-Based Recommendation Systems