PropertyMatch

Problem

Property search is usually just filtering (price range, bedrooms) — it doesn't learn what actually makes a property a good match for a given buyer's stated preferences versus what they end up interested in.

Architecture

FastAPI backend serving a Next.js frontend, PostgreSQL for property data, Redis for caching, and a Random Forest model trained on the property dataset doing the actual match-ranking rather than plain filters.

Key decisions

• —Random Forest over a simpler rules-based scoring system — the dataset has enough real features (price, location, size, condition signals) that a trained model outperforms hand-tuned weights.