Divyansh Rastogi

Experience

• 

  CERN

  Jun 2016 - Jun 2016

  CERN CMS Research

  Gained insights into antimatter production, Large Hadron Collider and Alice particle detector.

• 

  Max Planck Institute for Informatics

  May 2021 - Sept 2021

  Research Intern

  Working on an optimization platform used for modeling and solving mixed-integer optimization problems under Dr. Andreas Karrenbauer.

• 

  Indraprastha Institute of Information Technology, Delhi

  Aug 2021 - Apr 2023

  Working under Dr. Ranjitha Prasad on Personalized Federated Learning & Federated Self Supervised Multitask Learning Development Club of IIIT Delhi Competitive Programming Club of IIIT Delhi Worked under the guidance of Dr. Saket Anand on High-Resolution Satellite Imagery for Agriculture in collaboration with Google AI for Social Good and MSSRF.

  • Undergraduate Machine Learning Researcher

    Jan 2022 - Apr 2023

  • Byld Core Member

    Feb 2021 - Apr 2023

  • Foobar Member

    Oct 2020 - Apr 2023

  • Undergraduate Machine Learning Researcher

    Aug 2021 - Dec 2021
• 

  Scaletorch.ai

  Jan 2022 - Apr 2022

  Software Engineer

  Founding Software Engineer responsible for building and maintaing core components of the product.

• 

  Google

  May 2022 - Aug 2022

  Software Engineering Intern

  - Google Search for Apps team- Developed & tested an automation for tracking & debugging dropped documents in a search & index debugging tool through various stages of the apps search stack to optimize quality related search debugging.- Consolidated debug information of document signals, scores & ranks across the apps search backend.- Developed techniques to identify anomalies post document dropping & integrated overrides to gather further meaningful document signals & ranks. Show less

• 

  Scaletorch.ai

  Sept 2022 - May 2023

  Software Engineer

  - Founding Software Engineer responsible for building and maintaing core components of the product. - Designed and built the Distributed Asynchronous Processing engine (DAP) for increasing training throughput up to factors of 10X, by identifying training bottlenecks, leveraging optimizations through compute offloading, enhanced networking, and effective GPU utilization. The engine requires zero code change in the original client source code and is developed as a patch over PyTorch.- Designed and developed a zero code change, cloud-agnostic, encrypted & concurrent data streaming service as a virtually mounted filesystem supporting native file system traversals & operations over any cloud storage (even within archives) with event-driven LRU caching mechanism supporting local & global-tier-2 caching at Minio servers, drastically reducing cloud egress costs upto 60%. Using performant caching and listing solutions, the filesystem was able to emulate 1.1 million files in less than 2 minutes.- Set up efficient container orchestration, monitoring, and setups across multiple hosts for any client training environment, dependency management, custom indicators for health checks, orphan management, facilitating training and monitoring across 100s-1000s of hosts in parallel. - Built a custom, consistent, and minimal overhead logging service for viewing and traversing real-time training logs.- Set up and designed a centralized and dynamic reverse proxy abstracting user training artifacts / logs / results. Show less

• 

  Palantir Technologies

  Jul 2023 - Nov 2023

  Software Engineer

  Making Foundry's key ontology layer performant.

• 

  Google

  Nov 2023 - now

  Software Engineer

  Integrating new forms of payments.