Writing
Notes on making AI workflows reliable.
Short drafts focused on fuzzy evals, workflow boundaries, guardrails, and the gap between prototype success and deployment readiness.
How I Evaluate LLM Workflows When Correctness Is Fuzzy
A short note on separating format correctness from judgment quality, combining deterministic checks with human review, and making uncertainty visible.
From Prototype to Reliable Workflow
A practical checklist for moving an AI idea from impressive prototype to workflow component that can survive real usage.