May 2022 – Oct 2024 · 2 yrs 5 mos · Bengaluru, Karnataka, India · Hybrid

• Project: Real-Time Object Tracking and Smart Recognition on Edge Devices
• Overview:
• Developed a lightweight, intelligent object tracking algorithm optimized for low-FPS camera feeds on edge devices. The solution addresses key challenges in intrusion detection, vehicle/person tracking, and automated recognition where traditional state-of-the-art algorithms like Deep SORT, ByteTrack, and CenterTrack underperform due to their dependency on high FPS (>25 FPS).
• Key Highlights:
• Designed a tracking model capable of operating effectively at 3–5 FPS, focusing on near-field object detection (within 30 feet).
• Significantly improved event accuracy and alerting mechanisms for security systems by assigning consistent IDs across frames.
• Enabled real-time entry/exit counting and enhanced recognition accuracy for Automatic Number Plate Recognition (ANPR) and Facial Recognition (FR) under edge constraints.
• Additional Modules Built:
• Lightweight License Plate Recognition (LPR):
• Developed a compact LPR model optimized for edge deployment with high accuracy and minimal compute overhead.
• Camera-Based Parking Management System:
• Implemented a smart vision-based system to automate vehicle detection, slot occupancy monitoring, and parking event tracking.
• Facial Recognition-Based Attendance System:
• Built a contactless attendance module using FR integrated with camera feeds, ensuring real-time identity validation and log maintenance.
• Technologies & Skills:
• Python · Convolutional Neural Networks (CNNs) · OpenCV · Deep Learning · Edge AI · Object Detection · Tracking Algorithms

Mar 2020 – May 2022 · 2 yrs 2 mos · Bengaluru, Karnataka, India · On-site

• In Swiggy I majorly worked on two projects and both get published on Swiggy official platform on medium
• Project 1: Smart Business (Real time detection and explanation of business anomaly)
• Summary :
• Food delivery is a complex hyperlocal business, spread over thousands of geographical zones across India. Here zones represent smaller geographical areas. Every day, our business team is tracking and reacting to metrics that monitor the health of the business. These metrics are first level (L1) metrics like Cost Per Delivery (CPD) by zone at different time granularities and multiple L2 and L3 metrics like Daily Incentives, Minimum Guarantees, etc. which influence L1 metrics. Anomalous changes in these metrics need to be detected and corrective action needs to be taken to prevent potential business loss.
• we have used state of deep learning based forecasting algorithm along with regression based tool for causal analysis. Later this work is also used in Forecasting platform.
• I had executed this work with Viswanath (Data Scientist) and published blog in Swiggy official platform.
• https://bytes.swiggy.com/an-end-to-end-system-to-detect-and-explain-anomalies-in-operational-metrics-448bc74c700e
• Project 2: A real-time supply-shaping system to meet demand under constraints
• Summary :
• Maintaining optimal supply-demand distribution across zones in a city in real-time is an important aspect of the hyperlocal food delivery business. Here zones are smaller geographical locations inside the city and considered as Operations clusters. Delivery Executives (DE) are associated with a zone and all actions related to managing the gap between demand (orders being received) and the available supply of DEs are taken at the zone level. Gaps between supply and demand can occur due to a variety of reasons (Rain, Sudden spike in demand, competitor campaign etc)
• https://bytes.swiggy.com/a-real-time-supply-shaping-system-to-meet-demand-under-constraints-ff449a73874b

Jan 2019 – Feb 2020 · 1 yr 1 mo · Bangalore

• Detecting Vehicle Unscheduled Stop
• Developed Statistical Algorithm for Detecting Unscheduled Stop by Driver during the ongoing trip
• Developed Accept Reject Module for Trip to maximize Truck Utilization
• Come up with Novel Mathematical Formulation for Relay as a Service which maximize the Truck Utilization
• Pilot Dry Run Recommendation System
• Developed a Recommendation system to trigger the pilot dry runs aimed at reducing pilot dry run across Rivigo relay pitstop
• Pilot Recruitment Process
• Developed EM based algorithm using simulation platform which minimize Truck wait time and Pilot wait time across rivigo relay pitstop

Feb 2016 – Jan 2019 · 2 yrs 11 mos

• Location Intelligence
• Developed Bayesian based algorithm for Pick Up and Drop Location Recommendation, Also developed ML based Technique to identify location type
• Customer Intelligence
• Used Unsupervised algorithm to estimate customer regular rides and later this work is extended to compute Customer LTV and engagement score which was used by offer engine
• Guardian Platform for Customer Safety
• Developed Probabilistic Algorithm for Route Deviation and Unscheduled Stop
• Driver Scoring
• Machine Learning model is developed to generate Driver driving skill scoring, In this process we have processed IMU and GPS data to do road profiling and generated smoothness index,potholes and bumps on road. This scoring had very good correlation with scheduled maintenance cost of vehicle, Fuel efficiency and Accident.

Feb 2014 – Sep 2015 · 1 yr 7 mos · Bangalore

• Guest Profiling(Based on Inncom Hotel sensor Data)
• Research and Developed Large Scale Times Series Clustering for Guset HAVAC interaction and room sensor data. This data used for knowing predictive maintenance of hotel room, room allocation service and Guest sleep deprivation.
• Orion Energy Analytics
• Created Big Data Analytics Dashboard for HAVAC Sensor Data using Map Reduce Frame work

Aug 2012 – Feb 2014 · 1 yr 6 mos · Bangalore , India

• Forecasting System
• Researched and Developed a High Dimensional Forecasting System which scales for millions of seg-ment.
• eCPM Goal Seek
• Control Theory based solution for maintaining minimum eCPM for Publisher.
• Technology: Hadoop, Google Charting Library, R

Sep 2010 – Apr 2012 · 1 yr 7 mos · Gurugram, Haryana, India

• Hadoop Subsystem
• Developed hadoop based system for storing and processing Large scale Web Event Data for generating various SEM based Reports and also Data is stored in Various HBase table for Real time Data Access. Technology: Java, Hadoop, HBase, Linux
• Alerting Framework
• Developed Real Time Alerting Framework running over HBase to monitor and generating real time alerts for various processes running over hadoop cluster.
• Technology: JavaQuartz, Hbase, Java, Linux
• Bid and Budget Optimization for Social Network
• Researched and Developed the Bid and Budget optimization Algorithm for Facebook online Ads which takes care of AdRotation, AdFatigue, Pausing and Unpausing of Ad.
• Technology: Perl, R, Excel, JavaScript, Linux
• Bottleneck Analysis for SEM(Search Engine Marketing) Researched and Developed a novel algorithm for determining the Various AD Performance bottleneck for search engine online Ads
• Technology: Perl, R, Excel, JavaScript, Linux

Nov 2006 – Dec 2009 · 3 yrs 1 mo

• Network Anomaly Detection and Classification
• Researched, designed and implemented a data mining algorithm to classify various network anomalies such as network scan, port scan, udp floods etc. I developed a novel algorithm by combining a semi supervised classification algorithm and rule-based classification method to achieve more than 90% accuracy. Entropy based features were used for the classification.
• Technology: C, Matlab, Linux
• Sliding SVD
• Researched and developed a novel algorithm to update the existing SVD model in the streaming fashion. It avoids the re-computation of SVD for the slight modification in the data such as addition, deletion or modification of a few rows/columns.
• Technology: C, Matlab, Linux
• Nonlinear Modeling of Network Data
• Developed and Implemented Kernel methods for finding trend and outlier in time-series data.
• Technology: C, Matlab, Linux

Jul 2004 – Dec 2005 · 1 yr 5 mos

• Unicorn Development Language
• Implemented LL(k) based GRAMMER for Rule based language(UDL) which support data type and sql binding, In this project also implemented the Parser and Type checker
• Technology: ANTLR, Java