In-depth case studies from work currently in production. Each write-up covers the architecture, the design decisions, and the operational results. Each links to a detailed technical appendix where applicable.
A browser-native interactive diagram that walks through the agentic system running my job search. Sixteen specialist skills, scheduled jobs, Microsoft Graph wiring, and the data pipeline that keeps the tracker, dashboard, and digest in sync. Self-contained HTML with a click-to-play animation; opens in a new tab.
Open the interactive walkthrough →A scoped, shippable agent that triages inbound Outlook messages against a six-category classifier, drafts replies in the user's voice, and writes every decision to an audit log. Built on Copilot Studio fronting Azure AI Foundry, with Claude Sonnet via Foundry handling classification and drafting, Microsoft Graph for inbox, calendar, Teams, and Planner context, and Application Insights for the audit trail. The case study covers the tool catalog, the classifier, the voice profile, the data model, and the security posture. A technical appendix documents the Graph scopes, typed tool signatures, Dataverse schema, and Kusto audit queries.
Over the past ten weeks I designed, deployed, and have continuously operated an agentic AI system on Anthropic's Claude Agent SDK. It orchestrates Microsoft Graph, scheduled automation, analytics, and document generation across an end-to-end workflow. The case study walks through the architecture, the design decisions, the guardrails, and what it means for Microsoft cloud customers rolling out Copilot Studio and Agent 365.