A case study from a healthtech startup on deploying a production ML system for Parkway Pantai, Southeast Asia's largest hospital group. The system estimates patient bills at pre-admission, reducing mean absolute error by 55% and RMSE by 60%. Key insights include that ML accounts for only ~20% of the total effort, with data engineering, DevOps, and infrastructure making up the rest. The talk also warns against overfitting to batch/CSV data and Kaggle-style ensembles that don't translate to real production environments.
