A real-estate platform for buyers, sellers, and agents, with an AI co-pilot that analyzes disclosures, offers, and listings. I built it solo, front end to infrastructure.
Roosthaven is a real-estate platform that keeps buyers, sellers, and agents in one workspace from search to closing. Its AI co-pilot, Roost, analyzes the deal’s actual material: disclosures, contracts, inspection reports, and offers. I built the platform solo between December 2024 and April 2025, every layer of it.
The front end is Next.js. I scaffolded the first version with v0 and then reworked it by hand into a production interface, which was the right trade for a solo project: the tool got me a working UI in days, and the hand-tuning made it mine. The backend is Node.js, with NeonDB (serverless Postgres) as the data layer. The AI features sit behind the API and are fed by the platform’s own data: listings, tours, offers, and the documents people upload.
It runs on AWS, deployed through a GitHub-based CI/CD pipeline, so I could iterate alone without manual release steps.
My part wrapped in April 2025, and the platform is live at roosthaven.com with real Bay Area MLS listings. It has since been exhibited at C.A.R. REimagine 2025, the California Association of Realtors’ innovation conference.
I built every layer: the Next.js front end, the Node.js APIs, the database, the AI logic, and the AWS setup. Working alone forces honest calls about where to spend time, which is why the UI started in v0 and the database is managed.
An AI opinion about a house purchase is only worth something if it's anchored to the deal's real data. Roost works from the transaction's own documents, disclosures, offers, and listing data rather than answering from general knowledge.