We introduce the architecture of a technology-agnostic workflow engine that defines human-readable decision rules in YAML, ingests them into a RAG datastore, and leverages an LLM to retrieve relevant rules and instantly determine and execute the next approver.

First, we’ll demonstrate the end-to-end flow from rule definition through prompt design to datastore registration. Then, we’ll share production-ready best practices for maximizing retrieval accuracy, using GenAI to extract structured request data from unstructured documents, and keeping workflows current.

Attendees will leave with guidance on expressing approval workflows, practical techniques for structuring decision rules for optimal retrieval, and a roadmap for embedding AI-driven innovation into Drupal or any platform.