- Worked on the Ideation of Health and Wellness/ Fitness tracking platform Svastha.- Founded Svastha with the vision of establishing a fitness profile for every human, with aggregated metrics over time and provide the ability to preserve various fitness activity history [walking, running, cycling, yoga, hiking, cardio etc.] in one app.- Designed functionality to measure consistency in fitness activity in the form of streaks/badges and reward it appropriately by subsidized membership rates or discounts in fitness classes/fitness products etc

- Avala AI is a seed-stage startup, building a Multi-modal data labelling platform [image, video, LiDAR data] to create and manage high-quality training data for machine learning models.- Architected and built a Real-time Inference server in Python for Segment Anything Model, a computer-vision based 2D Image Segmentation algorithm to generate segmentation masks for the images.- Implemented auto-provisioning and de-provisioning of AWS Sagemaker instances with terraform- Benchmarked and iteratively optimized the inference server to sub-second response latency with GPU hardware accelerator on AWS SageMaker with pre-built Pytorch docker images- Designing and building a Model assisted labelling platform with support for various pre-trained Computer Vision models for Object Detection, Image Segmentation [Segment Anything Model, YOLO8, Depth Anything Model etc]

- Helped maintain decision engine, which is a soft-realtime distributed inference pipeline for processing shopper’s video streams from brick and mortar retail stores for seamless cashier-free autonomous checkout experience- Designed and implemented Redis Memorystore cache to persist the shopper to shard id mapping load balancer states to add the ability to scale the number of decision engine load balancer processes- Helped enhance the scalability/availability of decision engine service and in productionizing machine learning models- Implemented a startup buffer to minimize the message drops during service startup/redeploys- Implemented store layout parser python library for shelfs, shelf-sections, walls, ceilings, NFC reader, Exit etc. from LiDAR scan of the retail store and updated the down-stream services to consume the new layout- Contributed to the migration from frame number based processing of video stream to timestamp based processing to add flexibility to frame processing rate per second- Redis load balancer re-write to process lightweight Pose2D / Pose3D/ Checkpoint messages of shopper from the upstream tracking service- Enhanced service reliability by adding persistence in managed Redis Memorystore instances via periodic snapshotting to disk/append-only file- Leading exploratory effort to guarantee ordering of messages in Google cloud platform’s Pub-Sub queue- Enhanced Integration test suite to consume recent test data from Google cloud storage bucket to produce realistic results- Migrated CI/CD pipeline from Phabricator+Jenkins to Github Actions- Periodically triaged decision engine production service alerts/exceptions in Sentry- Roll-out/Deployment of the new features to more than 20 autonomous checkout retail storesTechnologies: Python, Redis MemoryStore, Protocol Buffers, Google Pub/Sub, Google cloud Platform

- Contributed to Konvoy, a managed Kubernetes platform to automatically provision kubernetes clusters on cloud and on-premise- Integrated Gatekeeper Open Policy Agent (OPA) as an add-on to Konvoy, to enforce organization wide security policies- Triaged Konvoy CI test failures and added pre-flight checks to verify if required modules are successfully installed on the provisioned machines in AWS

- Worked on a research project aimed at building a hypergraph analysis system. This project was built on top of Apache Spark GraphX library- Researched on the effect of batch size on the performance of streaming workloads and evaluated an algorithm that automatically adapts the batch size based on the data ingestion rate/workload of the system- Graduate Teaching Assistant for course CSCI 2021 - Computer Systems Organization- Worked on a prediction algorithm for environmental protection project which predicts if a chemical compound is environment-friendly or pollutant.

- Uber for Business - Backend/Platform Team- Built a new service 'Dayrunner' to enable monthly billing scheduling preferences for all of Uber's business related trips.- Also implemented warning, account suspension and restoration logic in case an organization failed to pay off their monthly bill.- Used Thrift to create language agnostic service interfaces and to maintain a scalable codebase in Uber’s service-oriented architectureTechnologies Used: Python, Thrift, Tornado(asynchronous web and networking framework), Celery, Kafka, Redis, Sentry, Vagrant, Jenkins, Git

- One Kings Lane is an online e-commerce platform for home interiors, furnitures and accessories with expert design services- Integrated One Kings Lane Studio Scheduling application into One Kings Lane website - Implemented duplicate elimination logic for recommended products in the product details page - Fixed existing bugs to enhance user experienceTechnologies Used : Ruby on Rails, HTML5, Susy Framework, Docker, Git, Jenkins

- Athenahealth is a cloud-based healthcare provider platform which maintains HIPAA compliant Electronic Health Records, appointment history of the patients, automatic claim creation and electronic remittance advice processing services for providers- Created robust and scalable workflow for processing ERA (Electronic Remittance Advice) received from insurance companies in ANSI 835 format and reconciled against billed claims.- Developed custom functionalities for specific insurance payers and web interfaces for enterprise clients.Technologies Used : Perl, HTML, CSS, Javascript, Oracle, Perforce

- GlobalScholar is a k-12 school education software company which helps maintain student information, grade books etc. with support for periodic state reporting to the school district- Developed a utility to automate the process of build and deployment that saved >50% of manual effort. -Also developed build verification tool that continuously integrates, builds and tests latest code commits.- Generated reports for periodic state reporting to the school district