Automated Valuation Model & Real Estate Bot

The problem

The client wanted two things for the Australian real-estate market: a model that predicts a property's price as closely as possible, and a bot that evaluates a property the way an experienced broker would, to find the deal that yields the highest profit. Neither existed, and both depended on property data that first had to be collected.

What I built

• A data-collection layer that pulled listings through APIs such as the Domain.com.au developer API and scraped additional data directly from real-estate websites.
• Feature engineering over the collected listings to turn raw property records into model inputs.
• An XGBoost regression model, chosen after comparing it against other algorithms on the same data; it outperformed the alternatives.
• A prototype of the broker evaluation algorithm, implemented from the client's description of how a broker assesses a property for maximum profit.

How it was delivered

Stage one was the valuation model. My target metric was R², and the final XGBoost model reached around 84%, a strong result for this use case. Stage two was the broker bot: the client shared the evaluation logic, I built the prototype and started testing. The alpha has all the features implemented; what it still needs is enough data to cover every combination of inputs the algorithm can ask for.

Outcome

An automated valuation model at around 84% R² for Australian property prices, built on data I collected through the Domain.com.au API and web scraping, and a feature-complete alpha of the client's profit-maximising broker bot. The remaining work is data coverage rather than code.