I build AI-powered web products with typed, auditable backends. Open to remote contract or full-time work.
Veris — GitHub PR analysis tool that classifies risk and surfaces a structured merge recommendation with evidence. GitHub App ingests PRs via webhooks; diffs go to an LLM through OpenRouter; output is schema-validated before it's ever shown to a user, not rendered directly. Can reliably check structural consistency, but treats true intent verification (did the PR solve the actual task) as an open problem rather than faking it. MVP, live: veris-roan.vercel.app
Briefr — AI research agent, cited reports in seconds. Citations pulled from grounding metadata, not generated by the model. Live: briefr-ai.vercel.app
AppealMD — turns dental insurance denials into payer-specific appeal letters in under 30s. 3-step pipeline (classify → match rule → generate), Pydantic validation at every stage. No PHI stored. Live: getappealmd.vercel.app
I avoid abstractions I can't audit. I make product decisions, not just implementation decisions.
Email: fraolashebir84@gmail.com