My House Fax

The challenge

Ryan Jones had already shipped one product with us. He came back with a bigger idea: a car has a service history that follows it to resale, and a house — the far larger asset — has nothing. Owners spend tens of thousands on a roof or an HVAC system and arrive at closing with a shoebox of receipts and no way to prove any of it moved the number.

The technical problem sits underneath that. Public property data is real but patchy, sometimes duplicated, occasionally wrong, and it reports the municipal assessed value rather than what the house is worth. A product that just resells that data at a markup has no reason to exist.

The approach

The baseline comes from Realie.ai — beds, baths, lot size, construction type, tax and assessment records, transfer history. Federal hazard data is layered on from FEMA, joined on the county FIPS code rather than the address, because FEMA’s index doesn’t resolve any finer than county. The proprietary layer is the owner’s own ledger: dated maintenance and improvement records, cost, contractor, receipt photos — the part no API can supply, and the reason anyone would trust the report.

The interesting engineering was in the gap. The property API reliably fails on new builds — a two-year-old house in Texas simply isn’t in the record yet — which dead-ends onboarding at exactly the wrong moment. We built a Gemini fallback, and the discipline was in what we didn’t ask it for. Early versions hallucinated because the prompt demanded deed transfers and tax assessments, data an AI has no honest way to know. Constrained to strictly public, observable facts and returning strict JSON mapped to the existing schema, it became accurate and fast. It also turned out not to be bound by the primary API’s US-residential limits: the same fallback returned good data for a commercial warehouse and for a property in Canada, opening international and commercial real estate at essentially no additional development cost.

Elsewhere, the same instinct — solve it at the seams rather than build a monument:

• Duplicate API records for a single address are filtered programmatically on the presence of valuation data, instead of asking a homeowner to pick the right one.
• The AI property assistant carries a rolling summary that recompresses context every 30 messages, so a long conversation doesn’t quietly become an expensive one. Its legal boundaries — no tax or legal advice, no guaranteed valuations — are hard-coded into the system prompt rather than left to tone.
• The printable report is a formatted web view that triggers the device’s own print-to-PDF, not a custom PDF generator. Same output for the user, a fraction of the build.
• Ownership verification is graduated: the first person to claim a property walks straight in, and only a second claimant has to upload proof. Gemini reads the document; a rejection asks for a different one before anything reaches a human appeal.
• Subscriptions run through RevenueCat, but the enterprise tier deliberately doesn’t. High-ticket brokerage deals are invoiced off-store and unlocked with a database flag — which keeps 15–30% of every one of them with the client instead of Apple and Google.

Where it stands

The web app is live at myhousefax.io and the mobile builds have been through internal QA and store-review preparation. Public launch is deliberately held until two remaining blocks land — a single verified ledger per property, and the notification system — because shipping first and merging property records afterwards is the kind of database problem that doesn’t get cheaper with time. We’ll update this page when it’s live.

See My House Fax: myhousefax.io