Building Intelligent Applications with Graph-Based RAG on PostgreSQL
summary
Demonstrates a legal Copilot built in Postgres on a Washington subset of the US Caselaw dataset (~100k cases, ~500k pages), converting opinions to embeddings with pgvector and azure_ai, adding graph relations via Apache age, and indexing with pg_diskann. Claims DiskANN shrinks index size (150 MB vs ~3.5 GB — ~25×) for faster, cheaper ANN at scale. Basic vector RAG yields ~40% recall on a prominence query; combining vector search, graph re-ranking and reciprocal rank fusion (GraphRAG) raises relevant-case rankings substantially, producing more accurate retrievals for natural-language legal queries.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:47Agenda / Why Postgres
- 3:47US Case Law Dataset
- 4:32Semantic Search
- 5:37Vector 101
- 8:25Basic RAG
- 9:09Common Problem with RAG
- 13:45Advanced RAG Architecture
- 20:40Resources
related talks
26mProduction RAG at Scale with Azure Database for PostgreSQL
33mVector Databases in the Age of AI: Embracing RAG for Enhanced Search Capabilities
39mEnhancing RAG with Neo4j Knowledge Graph
30mGraph RAG: Bringing together graph and vector search to empower retrieval
36mUnleashing the Power of Vector Databases for AI-Driven Applications
35mThe Hidden Jewels to Solve Confabulation Challenges in Generative AI’s RAG Promise
34mAnalytics and AI with PostgreSQL - Mission Possible!
24mFrom Queries to Agents: The Next Era of Data Retrieval on PostgreSQL