Rebel Foods

Predict Estimated Time of Arrival: Used feature engineering and tree based models to predict the ETA when a customer orders food from the app and dynamically updates it. Designed an AB test that showed the model decreased RMSE by 5% and reduced WIMO (Where is my order) by 10%. Also used interpretation models like LIME and Shapley and led calls to help business teams understand black-box models better. Optimized and deployed the model in production using AWS ECS,Docker, Redis, Flask and… Predict Estimated Time of Arrival: Used feature engineering and tree based models to predict the ETA when a customer orders food from the app and dynamically updates it. Designed an AB test that showed the model decreased RMSE by 5% and reduced WIMO (Where is my order) by 10%. Also used interpretation models like LIME and Shapley and led calls to help business teams understand black-box models better. Optimized and deployed the model in production using AWS ECS,Docker, Redis, Flask and Elasticsearch.Retention Optimisation: Developed an ML based strategy that predicts whether a customer will order in the next month based on factors like days since last order, CX of the last order, engagement on the app etc and then send the right communication to the customers who won’t. AB tested this against a rule based logic run by the business team. Increased retention by 20% while also increasing profits by 10%.Push Notification Personalization: Personalize PNs to customers by personalizing different components like the restaurant, discount and time. Lead multiple proposal calls to pitch this idea to the business team. Lead and mentored team members in problem formulation, data extraction and deciding the metrics. Ran multiple AB tests with iterations. This led to 10% increase in CTR and 5% increase in conversion.Customer contacts Forecasting - Developed and deployed an ML model which forecasts the count of contacts for a given day at an hour level, across contacts categories. This forecast is a key input for workforce planning and optimization. It helped in increasing workforce throughput by 5% as well as reducing avg wait-time by 9%.