The San Francisco startup builds voice-powered AI agents that take a single property address and hand back a finished contract packet - the deed lookup, the tax pull, the forms, all of it. Its bet: real estate software should do the work, not just store it.
Ask any real estate agent what eats their week and they will not say showings or negotiations. They will say the deed they had to hunt down at the county registry, the tax assessment they retyped into a listing agreement, the same street address they keyed into six different forms before lunch. RealPact, a three-person startup in San Francisco, was built on the belief that a computer should be doing all of that - and that, oddly, most real estate software still doesn't.
The company's one-line thesis fits on a whiteboard: real estate software should do the work, not just store it. Dropbox for closings, digital filing cabinets, e-signature routers - the industry has plenty of tools that hold documents. RealPact's argument is that holding a document is the easy part. Finding it, reading it, and filling out the next form from it is the part that quietly consumes an afternoon. So the company built AI agents to do that instead.
The demo is disarmingly simple. You give RealPact a property address. Behind the scenes, its agents fan out to county deed registries, tax assessor records, parcel databases and MLS listings, pull what they find, and assemble it into a draft: a listing agreement, a purchase contract, the compliance attachments, the deadlines. Work that used to mean an afternoon of tab-switching arrives as a packet in minutes. You can even start the whole thing by talking - describe the deal out loud, and the voice interface converts it into a structured document.
RealPact splits that job across four specialized agents, which is where the product gets its shape. Each one owns a slice of the transaction, and together they cover the paperwork spine of a home sale.
Retrieves deed records from county registries and flags encumbrances tied to the property.
Extracts tax assessments and figures from assessor documents using computer vision.
Consolidates ownership, parcel and property details into a single clean record.
Auto-fills contracts, attaches compliance docs, manages signatures and deadlines.
Real estate software should do the work, not just store it.
- RealPact's founding thesis
RealPact comes from Ranvir Deshmukh and Erik Peterson, two Dartmouth graduates who had already shipped something people actually used. Their first project, CourseMe, was an AI course advisor that reached more than 4,000 monthly active students at Dartmouth before it was acquired. That experience left a mark on how they picked their next market: not the flashiest problem, but a daily one, in a field where the incumbent software is old and the pain is specific.
They started RealPact in 2025 - the legal entity is registered as SignPact, Inc. - and built it out of New Hampshire before landing in San Francisco. In May 2026 they won Dartmouth's Magnuson Startup Competition and its $20,000 prize. A month later they announced a pre-seed round and a spot in Y Combinator's Summer 2026 batch.
RealPact didn't ship as a landing page with a waitlist. It went live inside real brokerages closing real deals. Its early roster includes Four Seasons Sotheby's International Realty, Black House Real Estate and Next Homes - the kind of names that suggest agents were willing to run a three-person startup's software on transactions that matter. The company began with brokerages in New Hampshire and is expanding from there.
That customer profile also explains the business model. RealPact sells to brokerages and their agents as B2B software, starting with design-partner and pilot accounts, with the stated ambition of growing from transaction paperwork into a broader operating system for how a brokerage runs.
One address. Every deed, tax, and property record pulled instantly.
- RealPact product tagline
The real estate transaction stack is not empty. Agents already have Dotloop, SkySlope, zipForm and DocuSign-style tools. But RealPact draws a line between software that stores and routes documents and software that retrieves records and completes the work. Most of the incumbents live on the first side of that line. RealPact is betting the second side is now buildable, because AI agents can read a county PDF, understand a tax assessment, and fill a form the way a junior assistant would.
Illustrative comparison of the paperwork burden RealPact aims to remove.
If there's a moat here, it probably isn't the language model - everyone in the YC S26 cohort has access to those. It's the plumbing: the connections into deed registries, assessor portals and MLS feeds that make the demo work on a real address in a real county. Those integrations are unglamorous and slow to build, which is exactly what makes them defensible.
RealPact lands in the middle of a broader wave: vertical AI startups that pick one document-heavy workflow and automate it end to end. Real estate is a natural target - high transaction value, heavy paperwork, aging software, and agents whose time is genuinely expensive. The company's own framing is that it's building an AI-native operating system for the modern brokerage, expanding its agents across the full transaction and, eventually, across brokerage operations.
For now the promise is smaller and more concrete, which is part of the appeal. RealPact isn't trying to reinvent how homes are bought and sold. It's trying to delete the afternoon you spend hunting for a deed. That is a modest pitch, and a specific one, and the founders seem content to keep it that way while the product earns its place inside more brokerages.