A real-estate directory that ranks itself
Search, enrichment, and ranking across thousands of listings — built to stay fast and accurate as the catalogue grows.
PurposX needed a real-estate directory that did more than store listings — it had to surface the right property for each query, enrich sparse data, and stay fast as the catalogue scaled into the thousands.
The challenge
Listings arrived inconsistent and incomplete. Naive keyword search returned noise, and relevance degraded as volume grew. The product lived or died on whether the first screen of results was actually good.
What we built
- Ingestion & enrichment — normalize messy listing data and fill gaps with structured extraction.
- Hybrid search — keyword plus semantic retrieval, so intent matches even when wording doesn't.
- Ranking — a tunable scoring layer combining relevance, freshness, and quality signals.
Engineering for scale
We treated relevance as a measurable target, not a vibe: a labelled query set and regression evals catch ranking drift before it ships. The data layer was designed for the catalogue to grow an order of magnitude without the first screen of results getting worse.
The outcome
A directory where search feels deliberate — the right listings, ranked well, returned fast — and a ranking layer the PurposX team can tune with confidence because every change is measured against the eval set.
More case studies
- How a PR agency scaled SEO and content with 30+ AI agentsA mid-size PR agency runs 30+ production AI agents — technical SEO, PR research, content at scale, and three niche directories — on a self-hosted runtime.
- Sensor-to-dashboard for vertical farmingA production ingestion pipeline — Raspberry Pi over mTLS, FastAPI, Postgres — feeding live dashboards and an assistant.
- A RAG assistant teams actually trustGrounded answers with citations, evals to keep them honest, and observability so drift gets caught before users do.