Graph RAG: Building Smarter Retrieval Workflows with Knowledge Graphs

The presentation discusses the advancement of Retrieval-Augmented Generation (RAG) by integrating knowledge graphs to enhance retrieval workflows.

The speaker, Cassie Shum, highlights several key points:

• Current Challenges with Traditional RAG:
  • Struggles with complex, multi-step, or domain-specific queries due to reliance on static pipelines and vector-only retrieval.
• Graph RAG Overview:
  • Utilizes knowledge graphs to incorporate entities, relationships, and provenance, improving retrieval quality and enabling traceable reasoning.
• Benefits of Graph RAG:
  • Allows the creation of agentic workflows using graph-native reasoning and declarative logic for planning and executing multi-step tasks.
  • Provides robust, intelligent retrieval workflows capable of handling complex data and queries.
• Practical Implementation:
  • The integration of structured and unstructured data into a knowledge graph allows for more refined queries and insights, as demonstrated with the Jaffle Shop example.
• Challenges and Solutions:
  • Emphasizes the need for clean and well-structured data as a foundation for successful implementation of AI systems.

The overall aim is to advance beyond traditional chatbots to develop explainable, automated systems that can adapt to new models, technologies, and AI developments.

This is the end of the AI-generated content.

Retrieval-Augmented Generation (RAG) has unlocked new capabilities for large language models (LLMs) by providing them with external context. But in real-world settings, many RAG systems still struggle—relying on static pipelines and vector-only retrieval that often fall short when queries are complex, multi-step, or domain-specific.

This talk explores how knowledge graphs can address these limitations, evolving RAG into a more structured and semantically aware system. We'll introduce Graph RAG, a practical approach that incorporates entities, relationships, and provenance to improve retrieval quality, enable traceable reasoning, and provide fine-grained control over what’s retrieved—and why.

Building on this foundation, we’ll share our work on extending toward agentic RAG: LLM-powered agents that use graph-native reasoning and declarative logic to plan and execute multi-step workflows. These systems go beyond passive retrieval and toward active problem-solving across structured data.

You’ll leave with practical patterns, architecture ideas, and an understanding of how to build retrieval workflows that are robust, intelligent, and ready for scale.

What you’ll learn:

• Where traditional RAG architectures break down in complex domains
• How Graph RAG leverages semantic context for better retrieval and reasoning
• How Graph RAG can leverage agentic workflows to enable multi-step workflows grounded in structured data
• Design patterns and architectures that you can apply today with graph-native systems
• How to build AI systems that go beyond chatbots and toward robust, explainable automation