Primary speaker:
Maurice Tam, Lead AI Developer & Systems Architect, MedIT - Temerty Faculty of Medicine
Additional speakers:
Philip Chung, Associate Director, Applications and Data Architecture, MedIT
Ian Nillas, Information Systems Developer & Integration Specialist, MedIT
Description:
MedIT manages many mission-critical enterprise workflows. Over time, workflow systems often accumulate years of institutional knowledge without sufficient documentation. When administrators change roles or retire, organizations can be left with complex automation environments that are difficult to understand, maintain or evolve.
This presentation introduces a practical methodology for building an AI-powered knowledge assistant using Claude AI to support workflow administration and documentation in enterprise environments.
The approach was developed to manage 179+ Laserfiche workflows across seven repositories at the University of Toronto’s Temerty Faculty of Medicine. The system translates workflow artifacts into structured Markdown documentation, connects them to Claude AI through MCP filesystem servers and applies a strict anti-hallucination guardrail framework to ensure the AI references only verified information. A central governance document defines role boundaries, prevents fabricated responses and maintains a traceable prompt interaction history.
We will introduce the operational challenge, the system architecture and the methodology for translating workflow artifacts into AI-readable documentation. Special focus will be placed on the anti-hallucination guardrail framework, which enables reliable enterprise use of generative AI for operational knowledge management.
A live Claude MCP and context-engineering demonstration will showcase an AI assistant (“Agent K”) answering workflow management questions using verified documentation sources. The demonstration will illustrate how AI can help administrators understand, troubleshoot, and document complex automation systems.
The session will conclude with a short Q&A discussion.
