AI-native engineer — I build and operate production AI systems. I own the architecture, the verification, and the production operations.
AI Engineer at The Best Notary (US legal services), where I built their document-automation platform and run it end to end — used daily by a 15-person operations team:
- LLM document-analysis pipeline (Claude, Gemini via OpenRouter) that classifies estate-planning documents and extracts signing requirements — with hallucination detection that verifies every extracted name and page number against the source PDF before anyone sees it
- Reliability & cost engineering: per-document cost caps, rate-limit-aware retries, watchdog jobs that recover stuck runs, per-run cost tracing in Langfuse
- 2,600+ automated tests, QuickBooks billing integration, and automated document intake from Make.com, Airtable, and Google Drive
- Stack: TypeScript/React · Supabase (Postgres, RLS, edge functions) · Trigger.dev · Cloudflare Pages
That repo is private (client work) — happy to walk through the architecture live.
I specify the contracts and invariants, agents implement, and everything ships through tests, evals, and production monitoring. The judgment layer is mine: what to build, whether it's correct, and when it's safe to ship. I've proven that judgment by finding and fixing real security vulnerabilities and cost bugs in systems that "worked."
TypeScript · Python · SQL · React · Supabase · PostgreSQL · FastAPI · Trigger.dev · Cloudflare · Claude API · OpenAI API · OpenRouter · Langfuse · LangGraph · sentence-transformers · Make.com · Claude Code · Codex
- linkedin-sourcing-agent — autonomous candidate sourcing: embedding-based ranking plus personalized outreach drafting
- crewai-solar-der-management & sunspec-adk-agent — multi-agent systems for distributed energy resource management (CrewAI, Google ADK)
ryash1867@gmail.com · LinkedIn · Remote, based in India with reliable overlap into US mornings


