Field notes: .NET architecture, applied AI, Azure and product engineering.
RAG became the default answer for every text problem, and four common shapes are cheaper, faster and more accurate when solved with the right tool instead.
Operational observability, not a pretty dashboard: three signals that matter, one correlation ID, sparse alerts, and an exporter that pays its rent.
A technical decision that locks in cost, latency and portability for years, framed by the three Azure-native options worth shortlisting.
MassTransit on RabbitMQ ships in an afternoon and produces duplicates in production by week two. The piece the default setup quietly leaves out.
OIDC in a monolith is a weekend. Across eight microservices it becomes a quiet design decision that defines the next two years of your platform.
The PoC was cheap. Production isn't. Four invisible cost vectors that quietly turn a clean Azure OpenAI bill into a board-level conversation.
When Blazor Server fits, when WebAssembly fits, and how to keep the surface area small enough to maintain.
A field note on golden sets, observability, and the parts of retrieval that are easy to overlook until they break the team's trust.
Three pricing modes I use on .NET and Azure AI engagements, and when each one wins.