Agents & automation

A working example of how I think agent systems should be built: boring, verifiable, and in daily use by its author.

The agentic CV pipeline

One master profile (schema-validated YAML with per-audience text variants) feeds a multi-stage pipeline: a job description is analyzed, content is selected and tailored by LLM agents under strict anti-fabrication rules (every metric must trace to a source document), then rendered through LaTeX to an ATS-safe PDF, checked, and logged to an application tracker.

Why it is on this page

Not because it is big - because it demonstrates the practices I argue for in production AI: a single source of truth, per-audience rendering from one content model, validation gates instead of vibes, and provenance for every claim. This website is built the same way.