I build backend systems the way a careful editor builds a sentence — one moving part at a time, each in service of the next. Scalable APIs, real-time data, and the unglamorous middle layer where most incidents quietly begin.
Lately I've been pulling generative AI into that work — MCP servers, RAG pipelines, agent graphs — and reminding myself that an LLM is another I/O boundary, with all the failure modes that implies.
A strong foundation in computer-science fundamentals means I can spot where AI-generated code is wrong and know exactly how to fix it. I use the model as a tool, not a crutch.