Data Engineering Podcast High Performance And Low Overhead Graphs With KuzuDB
Aug 18, 2025
Prashanth Rao, an AI engineer at KuzuDB, delves into the cutting-edge features of their embeddable graph database. He explains how KuzuDB tackles performance issues with innovative columnar storage and unique join algorithms. The conversation reveals KuzuDB's potential for enhancing graph applications, especially in edge computing and ephemeral workloads. Prashanth also discusses the growing interest in graph databases for AI integration and how Kuzu can seamlessly work with other data formats like Iceberg and Parquet.
AI Snips
Chapters
Transcript
Episode notes
Columnar Design Powers Large Single‑Node Graphs
- Kuzu is an embeddable, columnar graph database built from academic research and OLAP principles.
- It combines columnar storage, vectorization, factorization, and novel join algorithms to scale to billions of nodes on one machine.
Scalability Focused On Single‑Node Growth
- Kuzu defines scalability as seamless growth on a single machine by pushing operations to disk and using temporary files for intermediate results.
- This lets users handle hundreds of gigabytes to terabytes and billions of edges without changing workflows.
Combine Vector And Graph Retrieval
- Combine vector and graph retrieval to give LLMs better context in RAG systems.
- Use Kuzu's built‑in vector index alongside its graph retrieval for richer AI retrieval layers.