Engineering deep-dives on AI systems, .NET, Azure, and production software. No fluff — only things I've actually shipped.
Everyone's asking which model to use. Nobody's asking what they're feeding it. How I fixed inconsistent agent output by engineering context — not chasing model upgrades.
From Semantic Kernel plugins and hand-rolled session management to one-line agent wiring. What SK still left on the table, and what Microsoft.Agents.AI finally solves.
How I built a code-first feature flag system for trunk-based development — per-environment state, client-safe filtering, expiry enforcement, and live toggling without a single redeployment.
Building a production multi-tenant NL query system over live data — with prompt injection hardening, AsyncLocal tenant isolation, and 90+ automated security tests.
How I built a 20-agent pipeline that takes a story description all the way to a merged pull request — brainstorming, coding, parallel quality gates, and an AI agent that patches its own failing builds.