EvalSports is a pre-seed startup focused on employing AI to combine multiple feeds for a game to generate production-quality highlights for fans and present real-time statistics and strategic insights for coaches and managers- Designed and implemented critical sports analytics components of our pipeline like real-time homography computation, ball-possession graph creation, and team- and player-level statistics computation. Actively involved in in-game validation of these components- Involved in NDI validation and player tracking implementation in deepstream pipeline

- Curated and benchmarked large geo-spatial datasets as part of ML Team. This involved training on 32 A100 nodes with scalable data delivery and ingestion.- Led a team of 3 in design, implementation and deployment of major components of GeoEngine’s scalable MLOps pipeline. This entailed building from scratch, and scalable deployment of critical backend features like inference pipeline, experiment tracking, dataset export, project and dataset level statistics' generation, billing and order management, and automated location-based clustering of images for change-detection and multisource remote-sensing datasets.- Responsible for the development and deployment of crucial AI-based solutions like automated hail damage detection and roof structure detection for our roof analysis tool: ianalyze.properties- Involved in drafting and proofreading research papers and presentations for conferences

Worked as developer in Fusion Applications Life Cycle Management team- Our team develops and manages backend applications for monitoring and managing lifecycle of WLS and OPMN components, credentials management, data migration, and scaleout.- I was involved in end-to-end development of a number of lifecycle management projects and related microservices, which were widely consumed by provisioning, upgrade and patching phases of cloud lifecycle- Security point of contact for these lifecycle-related projects

I was part of a team that designed SOC and IP development methodologies to improve correctness and efficiency of various TI design groups- Automated development of programs to be run on IPs through generation of custom IP spreadsheets from IP specs where IP related programming sequences could be written and translated to C output through which we managed to reduce overall development time by 30%