House Numbers
Core engineer on a multi-service TypeScript monorepo for an AI-powered mortgage-lending platform, working across backend services, frontend, and infrastructure.
- Designed and shipped production LLM pipelines for document data extraction, including a fallback lane that lifted field accuracy from roughly 0.30 to 0.87 on documents an existing OCR vendor could not process.
- Built the platform's communication infrastructure end to end: asynchronous email sync with durable retries and reconciliation, a new SMS channel behind a channel-agnostic abstraction, and LLM-driven message classification and matching over encrypted data.
- Turned agentic AI features into engineered systems: evaluation harnesses with LLM-as-judge scoring and human-graded ground truth, per-call cost tracing, and version-controlled prompts.
- Led an evaluation of vision-LLM document classification and extraction as a potential replacement for the incumbent OCR vendor, making large multi-page closing packets processable at a fraction of the vendor's per-document cost.
- Designed and delivered a monorepo-wide, agent-consumable documentation system with automated integrity checks in CI, alongside broad ownership of shared packages, public borrower upload flows, an account-offboarding subsystem, observability, and production incident response.
- typescript
- nodejs
- react
- mongodb
- aws
- redis
- kubernetes
- openai-api
- anthropic-api