Data Scientist @CRED, Bengaluru, India
- Allocation Rule Engine: Developed scalable microservices for intelligent reward allocation in UPI payments, leveraging user’s churn score & real-time fraud/abuse detection to enhance engagement across 60Mn+ monthly transactions. Worked closely with product and business teams to execute experiments optimizing cash burn, ensuring high retention.
- Identified subtle inefficiencies in the data flow and rerouted allocation data via Kafka to S3, bypassing DynamoDB to cut latency by ∼7ms and save $70/month in write costs, enabling efficient downstream Databricks ingestion for Analytics.
- Distribution library: Developed an internal PyPI library for optimizers used across reward allocators. Salient features of the library include abstracted notions of rewards, state management in sequential scenarios, reducing config footprint, etc.
- Cashback Sensitivity Model: Developed a cashback sensitivity causal model to identify situational pockets in the user journey in the CRED app where they are sensitive to cashback. This model aimed at making smart decisions by relocating the cashback from sure-shots to persuadables.
July 2022 - Aug 2024
Data Scientist @Cult.fit, Bengaluru, India
- Exercise Rep Counting: Worked on an end-to-end vision-based exercise rep counter system used in smart mirror device used in Cult gym’s via a separate app. The system supports rep counting for over 100 exercises.
- Cure.fit's SOTA pose-estimation model. Applications of this model include: Cult’s Energy Meter, Rep Counting, etc. https://blog.cult.fit/posts/winning-interactivity-using-computer-vision
- cult live Recommendation System: Top Picks for you widget: recommending new DIY sessions based on the Trainer, Format, Duration, Intensity affinity of a user. Try Something New widget: recommending new formats (Dance, Yoga, HRX, etc.) for users to try based on other people's collaborative behavior.
- pdf-parser: users vaccination status checker (uses techniques like regex and OCR).
June 2020 - June 2022
Summer Intern @Weizmann Institute of Science, Israel
Automation of Biomedical Image Analysis. Supervisor: Dr. Ori Avinoam.
- The goal of my project was to live track the cells/nuclei i.e. to create an autopilot for smart Acquisition in a microscope with several advanced imaging techniques like TIRF-M, confocal microscopy, etc. for imaging the cell population.
- Made use of both classical and iterative Deep Learning strategies for image segmentation of the entire cell population and extracted features from the segmented images.
- Designed cost functions for global optimization to track the cells and also worked on advanced algorithms like Kalman Filtersand Active Contours method for tracking.
May 2019 - July 2019