When an AI agent denies your insurance claim, can it explain *why*? Not with a hallucinated summary, but with a traceable, auditable chain of reasoning grounded in formal logic.

Most agentic AI systems today are plumbing: retrieve chunks, call an LLM, hope for the best. This workshop takes a fundamentally different approach. Instead of bolting intelligence onto retrieval, we build it into the data layer itself using W3C standards (OWL, SHACL, SPARQL) and open-source tools, with AWS providing the managed infrastructure.

Participants will construct a complete semantic reasoning pipeline from scratch:

- From metadata to ontology — inspect a relational schema, understand how schema elements map to ontology constructs, and explore the resulting OWL ontology. We'll discuss ontology induction approaches, from schema-driven to LLM-assisted, and where automation is practical today
- Encode business rules as SHACL shapes — declarative, composable, auditable logic that derives insurance denial reasons without a single line of application code
- Query through the ontology, not around it — use Amazon Bedrock to translate natural language into SPARQL, with the ontology providing the vocabulary that constrains and guides the LLM
- Ground every AI output in graph evidence — denial explanations are generated from SHACL violation reports, not hallucination. The reasoning is formal; the communication is natural language
- Bridge the virtual and the persistent — query relational data as RDF through a Virtual Knowledge Graph (Ontop), load curated knowledge into Amazon Neptune, and see how the ontology serves as the integration contract across both
- Wrap it into an agent — every capability you built becomes a tool an AI agent calls autonomously: schema discovery, SPARQL generation, SHACL validation, and denial explanation. The ontology defines what the agent can do; the shapes constrain how it reasons

The result: an AI system where every decision is traceable to a formal rule, every explanation is backed by graph-derived evidence, and the ontology, not the prompt, is the source of truth.

Built entirely on open standards and open-source tools. AWS provides the managed primitives (Amazon Neptune, Amazon Bedrock, Amazon SageMaker) so you can focus on the architecture, not the infrastructure. You'll leave with a working system and a methodology you can apply to your own domain.