About

The Mortgage LLM is a housing finance information and decision support platform for mortgage professionals.

Our mission is to organize and make accessible the world’s collective knowledge of the US housing finance system.

We do this through two complementary surfaces, both anchored on one of the most comprehensive US residential-real-estate data stores assembled — billions of records spanning HMDA application disclosures, the GSE (Fannie + Freddie) loan-level performance sets, and Ginnie Mae’s monthly loan-level data, augmented with reference data on lenders, conforming limits, and Census geography.

Mortgage chat — for asking. Natural-language access to the underlying data. Ask anything — denial rates by program, FHA share trends, lender-specific origination patterns, GSE delivery mix by vintage, geographic flows — and get cited, structured answers in seconds. Built for exploratory analysis, market intelligence, regulatory research, and competitor benchmarking. No SQL required.

Scoring models — for deciding. A suite of pre-trained, isotonic-calibrated probability models for the loans you’re underwriting today: pull-through, credit denial, Higher-Priced loan flagging, Early Payment Default across 12 / 24 / 36-month horizons, prepayment curves, repurchase risk, GNMA-specific EPD, and Fannie-vs-Freddie channel choice. Each model returns a calibrated probability, a risk band, and an operating recommendation tied to its intended decision use — pricing, decline gates, reserve setting, or MSR valuation.

Chat is descriptive (“what’s happening, why”); the scoring models are predictive (“what should I do with this loan in front of me”). Pair them, and you have a closed loop from market intelligence to per-loan decisioning, anchored on the same underlying data.