Multimodal Graphs
Most AI implementations are stuck in a "text-only" world, relying on flat vector searches that miss the deeper context found in visual and auditory data. While RAG was a starting point, the real breakthrough lies in Associative Intelligence: the ability for a system to link disparate signals—a diagram, a spoken nuance, and a written report—into a navigable world model.
In this session, Amy Hodler and David Hughes explore the architecture of Multimodal Graphs. We’ll demonstrate how to treat images and audio not just as blobs of data, but as structured components of a graph that enable true contextual reasoning.
What We’ll Explore:
The Shift to Associative Intelligence: Why the next leap in AI isn't just better retrieval, but better association across different modes of human expression.
Decomposing Multimodal: Strategies for breaking down images and audio into graph-ready components that provide rich, non-textual signals.
Agentic Graph Construction: Using Boundary BAML to move from unstructured noise to validated graph data that agents can actually reason through.
Designing for Context: How to build graph schemas that allow AI to traverse relationships between what it "sees," "hears," and "reads."
The Goal:
This session is designed for Data Engineers and Scientists who want to move past the limitations of text-centric AI. We will provide a high-level architectural overview and practical design implications for building a more comprehensive, explainable approach to AI-driven insights.
Prerequisite Knowledge:
To follow along with the demonstrations, you should be comfortable with Python and have a basic grasp of graph and vector database concepts.