Knowledge Graphs, Kuzu, and Building Smarter Agents | S2E4

13 snips

Sep 24, 2025

In this conversation, Prashanth Rao, an AI engineer at Kuzu with a wealth of knowledge in graph databases, reveals the groundbreaking features of Kuzu. He explains how its columnar, strictly-typed architecture enhances query speed and scalability. Delving into knowledge graphs, Prashanth discusses how large language models (LLMs) streamline graph construction from unstructured data and enable seamless querying via natural language. Insights into integrating Kuzu with existing databases and the future roadmap of the technology make this a must-listen for AI and data enthusiasts!

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

Strict Schema Enables Performance

Kuzu enforces a strict typed DDL because its node tables are columnar, enabling compression and batch processing optimizations.
Knowing types ahead of time unlocks performance gains uncommon in typical NoSQL graph systems.

INSIGHT

LLMs Bridge Graph Build And Query

LLMs accelerate both construction and querying of knowledge graphs by structuring unstructured data and translating natural language to Cypher.
This downstream natural-language-to-Cypher capability reduces the barrier to access graph data for non-experts.

ADVICE

Constrain LLM Extraction With Schema

Use LLMs to auto-extract structured graph data from unstructured sources, then validate against a strict schema.
Constrain model outputs to defined property types so you gain insights without losing schema safety.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Kuzu, Knowledge Graphs, and the AI Revolution with Prashanth Rao

In this episode of the Hedgineer Podcast, host Michael Watson is joined by Prashanth Rao, AI Engineer at Kuzu, for a deep dive into the world of embedded graph databases and their pivotal role in the age of AI. Building on the previous week's discussion of the columnar database DuckDB, this conversation explores Kuzu, a parallel concept focused on creating high-performance, embedded graph databases.

Michael and Prashanth explore why Kuzu's unique architecture—an embedded, columnar, and strictly-typed system—is delivering incredible speed and scalability for complex analytical queries. They discuss the resurgence of interest in knowledge graphs, driven by the need to impose structure on data for modern AI and LLM workflows. Prashanth explains how LLMs are revolutionizing both the upstream construction of graphs from unstructured data and the downstream querying of these graphs by translating natural language into the Cypher query language.

Tune in to learn about:

What Kuzu Is: An in-process, embedded graph database designed from the ground up for query speed and scalability, blending the benefits of columnar processing with the property graph data model.
AI and Graph Synergy: How large language models (LLMs) assist in both building knowledge graphs through structured output extraction and accessing them via natural language to Cypher translation.
Modern Data Stack Integration: The role of Kuzu as a powerful secondary semantic layer that interoperates seamlessly with primary data stores in data lakes (like Parquet files on S3), DuckDB, and Postgres.
Programming with LLMs using DSPy: A detailed look into using DSPy to create structured, modular, and optimizable prompts for more reliable LLM applications, bridging the gap between deterministic code and fully agentic workflows.
The Enterprise Knowledge Graph: A discussion on how a knowledge graph can become the single source of truth for understanding complex organizational workflows, data lineage, and interdependencies across an entire enterprise.
The Future of Kuzu: A preview of what's next for the Kuzu project, including new graph algorithm capabilities and enhanced concurrency for reads and writes.

Whether you're a data engineer, AI practitioner, or a leader in the asset management space, this episode provides a masterclass on leveraging modern database technology to build powerful, scalable, and intelligent applications.

Hedgineer.io

Follow The Hedgineer Podcast:

YouTube: (https://www.youtube.com/@hedgineer)
LinkedIn: (https://www.linkedin.com/company/90976838)
Twitter: (https://x.com/hedgineering)
Instagram: (https://www.instagram.com/hedgineer/)

Hosted on Acast. See acast.com/privacy for more information.