I'm a backend engineer who drifted into data and AI because that is where the interesting failure modes are. Most of my work is the part users never see: pipelines that ingest messy external sources, APIs that stay boring under load, and LLM stages that are measured rather than guessed at.
I build and automate systems with AI in the loop — not as a demo, but as the way the work gets done. That means LLM agents wired into real engineering workflows, custom tooling and MCP integrations around them, and evaluation harnesses so a model or prompt change is an experiment with a number attached instead of an argument.
What I care about in a codebase: clear layer boundaries, business rules that live in exactly one testable place, and failures that are loud and cheap rather than silent and expensive.