Cloud Solution Architect at Microsoft. I work with enterprise customers on Azure platform architecture — mostly Azure and AI platform engineering, API management, and the cost-governance problems that surface once AI workloads stop being pilots and start carrying real traffic.

What I publish here is the reusable half of that work: the patterns that held up across more than one engagement, written down so the next team starts further along than the last one did.

Opinions and designs here are my own and do not represent the positions, strategies, or opinions of my employer.

azure-ai-gateway-cost-attribution — Production-grade Terraform for running Azure API Management as a centralised AI Gateway in front of Microsoft Foundry models, with optional per-application cost attribution and chargeback.

It addresses three problems that recur on nearly every engagement:

• A model name hard-coded into forty services. Clients call a logical alias, never a deployment name, so a model upgrade is a gateway change rather than forty pull requests.
• No defensible answer to "what did each team spend?" Per-request token accounting across input, cached, output and reasoning classes, allocated against authoritative Azure Cost Management spend — gateway telemetry decides who, Azure decides how much.
• A gateway that works until one region throttles. Native APIM backend pools with priority groups and per-backend circuit breakers: reserved capacity first, automatic spillover, automatic cross-region failover.

Includes architecture documentation, eight architecture decision records explaining why each trade-off was made, and an honest list of what it does not do.

Centralised AI gateways · Azure API Management policy engineering · Azure Infrastructure Architecture . DevOps and Automation . FinOps and chargeback for AI workloads · Terraform module design · Identity-based access over shared secrets