We're building agentic AI for mortgage lending — automating document processing and income qualification for loan originators.
Underwriting a loan means pulling numbers out of a few hundred pages of paystubs, W-2s, bank statements, and tax returns, then applying underwriting guidelines that run to thousands of pages of edge cases. Most of it is still done by hand.
Problems we actually work on: routing scanned pages to OCR vs. text extraction when the text layer is unreliable; extraction pipelines where a wrong number is much worse than a missing one; running long agent workflows durably so a failure three steps in doesn't restart the whole loan; and evaluating multi-step output against guidelines that are themselves ambiguous. We put real effort into eval infrastructure, because "the demo looked good" does not survive contact with a lender's compliance team.
Stack: Python, PydanticAI, Restate for durable execution, Postgres, AWS. Frontier models from OpenAI and Anthropic plus open-source models where they win on cost or latency.
We raised an $8M seed led by Kleiner Perkins and have paying customers in production. Team previously at Snowflake, PayPal, Twitter, Atlassian, and Microsoft.
Comp: $250k+ base, meaningful seed-stage equity, 5% 401(k) match, full health coverage, commuter stipend, gym membership, daily office breakfast & lunch. Full relocation to SF covered.
Onsite in SF. Visa sponsorship: yes.