Dec 2019 – Oct 2023 · 3 yrs 10 mos · Bengaluru · On-site

• Led a team of 6 SDEs to architect and deliver a real-time ML inference platform for live sports, orchestrating specialized CV detectors under a strict 60s end-to-end latency SLA.
• Developed execution wrappers and isolation layers for scientist-authored detectors, managing GPU memory allocation, dynamic batching, and containerized deployment to maximize throughput.
• Architected unified streaming pipeline supporting HLS (4s), DASH (2s), and variable-length DAI segments; integrated detector orchestration, per-segment aggregation, clip boundary detection, and encoding workflows.
• Engineered hybrid sequential and concurrent execution strategies to balance sliding-window model context requirements with live-edge drift constraints; introduced frame-skipping to maintain stability under peak load.
• Implemented checkpoint-based failover using segmentId tracking and heartbeat monitoring to enable rapid recovery for stateful detectors during live broadcasts.
• Introduced live-edge drift observability using detection completion lag metrics with automated mitigation controls to preserve SLA under load.
• Eliminated dependency on third-party replay vendors by enabling internally generated clips and real-time inference workflows, contributing to $2–3M annual cost savings.
• Architected end-to-end CV/ML pipeline powering X-Ray Replay features (Spotlight, NameTag), overlaying contextual visual graphics on replay clips to enhance viewer understanding.
• Designed ML + human-in-the-loop workflow where detector outputs were validated and refined using CVAT-based annotation tooling internalized for Amazon deployment pipelines.
• Built orchestration logic combining detector outputs and human feedback to generate final visual overlays rendered into replay assets.
• Partnered across ML scientists, video encoding teams, and product stakeholders to productionise replay enhancements across live sports events.

Mar 2019 – Nov 2019 · 8 mos · Gurgaon, India · On-site

• Senior Backend Developer at Search And Discovery team at Zomato.
• Led team of 3 SDEs owning Reviews and Ratings backend services handling 3M requests per minute ( 50K RPS).
• Migrated PHP monolith to Golang microservice architecture, improving throughput, latency, and operational reliability.
• Designed and launched Solr-based Reviews Search platform with ingestion pipelines, boosting strategies, and query expansion.
• Implemented read-after-write consistency mitigation using primary routing and short-TTL caching; resolved concurrency vulnerabilities in rating updates using Redis-based distributed locking (SETNX).
• Used ZooKeeper-based dynamic configuration for feature flags and A/B experimentation.
• Built highly performant search for reviews on restaurant page with filtering, sorting and query searching capabilities.
• Tech Stack: Golang, PHP, Js, MySQL, Redis, Memcached, gRPC, Protobuf, Docker, ECS, Solr, Envoy.

Jun 2017 – Feb 2019 · 1 yr 8 mos · Noida Area, India

• At ION, I worked in Algorithmic Trading Team as a Software Engineer (Backend). I was the sole developer for the flow-internalisation product which was being used globally by leading investment banks. Worked on various feature enhancements and gateway integrations. Developed generic tools in TCL to perform and automate performance benchmarks.
• Worked on enhancing algorithmic trading frameworks developed in-house by the team.
• Tech Stack : Core Java, C++, Maven, TCL, Python, Robot Framework

Oct 2016 – Apr 2017 · 6 mos · New Delhi Area, India

• I worked as backend developer intern at Ziploan. During my time at Ziploan , I mainly worked on core problems that they were facing which included parsing transactional sms to extract relevant financial information about a customer such as bank account , bank name , debit/credit amount, existing loan emis, defaulting of emis , total current balance, debit card number etc. This information is one of the factors that help in making lending decisions and recognizing possible defaulters.
• Tech Stack :
• Python / Django , MongoDB

Jun 2016 – Jul 2016 · 1 mo · Noida Area, India

• I worked as a Full Stack development intern at UrbanHunt , where I developed the company flagship product , Incent Social , a customer engagement platform.
• The product is a shopify app that allows the store owners to provide loyalty points and discounts to those customers which shared the image of the fulfilled products and shared them on social media accounts like facebook and instagram.
• Tech Stack :
• Python / Django , Celery , RabbitMq , HTML , CSS , JS , Heroku , Shopify Api , Instagram Api , Facebook Api

Jan 2016 – Mar 2016 · 2 mos

• I was a member of Stanford Crowd Research Initiative headed by Professor Michael Bernstein , which involved the development of a crowdsource platform known as "Daemo" .
• We worked on Developing the platform in Angular and Django.