myOS Real Estate
A FastAPI microservice that turns raw real-estate datasets into rankings, analytics and semantic search, built in a day for a Hub71 hackathon, with the LLM deliberately kept out of the arithmetic.
The brief
Abu Dhabi’s AI PropTech Challenge, one day, Cursor × eVoost at Hub71: turn a pile of real-estate datasets into something a property agent could actually query.
What it does
A FastAPI service ingests six synthetic datasets plus OpenStreetMap data and exposes property and district rankings, semantic search over neighbourhood amenities, and downloadable PDF reports. It also implements a knowledge-base adapter conforming to ProfSidekick’s /v1/agents/ contract, so it can sit behind another agent as a tool rather than as a standalone product.
The design choice that mattered
The project’s own framing: the maths is deterministic Python, the LLM only explains and routes. Rankings and analytics come from pandas scoring with no model in the loop; a LangChain agent over OpenRouter (Claude Sonnet 4.5) sits on top to answer natural-language questions and decide which deterministic endpoint to call, not to compute the answer itself.
What it isn’t
Not a finished product. It’s a one-day hackathon scope decision, not a launched service: deployment is Railway-ready, but built for a challenge deadline, not production traffic.