Automating ML across property listings at PropertyGuru

PropertyGuru handles a large volume of listings and wanted to use machine learning across several jobs at once: image moderation, churn prediction, credit forecasting, and measuring listing performance. It became clear early on that managing user data and retention, and keeping the app experience up to the company's reputation, needed ML. Doing it manually did not scale, so PropertyGuru needed to automate its ML process.

PropertyGuru found it could run its entire end-to-end pipeline on H2O Driverless AI. The team ingests images, churn, and tabular data, explores it with AutoViz to spot outliers and anomalies, and builds models fast using pre-existing recipes such as churn, plus automatic feature selection, feature engineering, and hyperparameter tuning. It then deploys multiple models straight into AWS Lambda from within Driverless AI, running them simultaneously as Java objects and watching their performance on live data.

The data science team iterates on new and existing models far faster, because Driverless AI handles the feature engineering, visualization, and tuning as new features are added. Non-technical teams can work with the data more easily, and the marketing team got a head start on churn prediction rather than building a model from scratch. With model building taking much less time, the team has room to innovate, including experimenting with Google Lens.