Jitesh Malipeddi

Collaborated with Product and Engineering teams to solve and engineer an end-to-end system based on DBSCAN based clustering, post-processing and integrations with downstream systems to enhance the accuracy of customer addresses, which led to CX improvement by reducing 100s of cancellations due to incorrect addresses and 1000s of unnecessary calls from delivery-partners to customers per day. As part of a drive to reduce reliance on 3rd party Maps providers, I built a pipeline to optimize distance serving using the Open Street Map Network instead of Google Maps leading to a savings of INR 1.5 lakhs per month while causing no adverse impact on check metrics. Revamped the existing method of ranking customer addresses in the app to move away from a heuristics-based legacy logic to a simpler, lightweight logistic regression model using intuitive features. This led to a 10% increase in the metric measuring ease of effort for customers. Developed an end-to-end fraud detection pipeline to flag delivery partners who were spoofing their GPS to exploit Swiggy’s assignment and payout systems. Worked with risk investigators and frontline Ops to understand the patterns and came up with a solution that involved DBSCAN clustering leading to abuse being caught at a 95%+ precision. Proficient in Spark (Databricks), SQL (Hive, Snowflake), Python, SDLC tools like Git and Jira in an AWS environment. Shipped multiple ML models on Swiggy’s DSP platform and owned their operational excellence.

Vehant Technologies is a specialist Computer Vision company focusing on Security and Traffic Enforcement Solutions. As an intern, I worked primarily on exploring object detection models keeping in mind the trade-off between accuracy and speed. Implemented knowledge distillation techniques enhancing the performance with a 7pp increase in Mean Average Precision.

Assisted faculty in conducting lab sessions and tutorials for the Data Structures and Algorithms course. Programming languages - C++ and Python.

Utilized deep learning techniques to perform Optical Character Recognition (OCR). Deployed the OCR model on the Nvidia Jetson Nano kit for real-time inference.