Jun 2022 – Jun 2025 · 3 yrs · Mountain View, California, United States

• Research Engineer working at Google Deepmind after the merger between brain and deepmind
• 1) Worked on the open source MLCommons AlgoPerf competition where folks from all over (Meta, Google, Nvidia etc ) submitted optimizer implementations for us to benchmark using our infra :
• competition code: https://github.com/mlcommons/algorithmic-efficiency
• paper: https://arxiv.org/abs/2306.07179
• 2) Collaborated on the seed project to push SoTA on transformers playing chess without using any search strategies : https://arxiv.org/abs/2402.0449

Apr 2021 – Jun 2022 · 1 yr 2 mos

• working in brain research on benchmarking training algorithms for large scale deep learning models, we use JAX :)
• 1) benchmarking neural network training algorithms
• 2) contributing to external benchmarks as part of MLCommons algorithms group

Jul 2019 – Apr 2021 · 1 yr 9 mos

• 1) Worked on YouTube Ads ML team owning conversion models for predicting in-app purchases and spend, developed optimizations to better handle extreme class imbalance, delivered modeling improvements for boosting NBU performance, implemented proper regularization and constraint mechanisms to improve generalization.
• 2) Developed infrastructure for LDA based unsupervised user embeddings for 2 billion users based on user watch history leading to improved performance in downstream supervised tasks.
• 3) Designed, implemented and launched changes to backend serving systems serving 1 million+ QPS.
• 4) Ideated and launched features to model user conversion behavior resulting in increase of revenue by 5% on corresponding traffic.

May 2018 – Aug 2018 · 3 mos · Sunnyvale, CA

• Worked with the Yahoo! Small Business team in their e-commerce division to integrate Authorize.Net payment gateway.
• Designed backend REST APIs (Java, C++) to support the integration.
• Integrated push notifications in online stores to increase traction.
• Developed a cli tool (ybuddy) to automate deployment.

May 2017 – Aug 2017 · 3 mos · United States

• According to Google's open source blog, this year 1318 out of 20,651 students were accepted.
• Reduced file parsing time from 2000ms to 100ms by writing Javascript parsers for genomic formats.
• Implemented algorithms to efficiently retrieve data from remote files containing up to 5 GB of genomic data within 6 file reads.
• project link: https://summerofcode.withgoogle.com/archive/2017/projects/5774476155289600/
• github repo: https://github.com/EnsemblGSOC/sourabhr-gsoc-2017

Apr 2015 – Jan 2017 · 1 yr 9 mos · Remote

• Level 2 seller on Fiverr.com, worked with 50+ clients from 15 different countries.
• Developed web apps, 2D games in Javascript, wrote Python scripts, performed testing using Selenium
• Worked on hardware projects using Raspberry PI and Arduino, and wrote image processing scripts using OpenCV
• Received 80 excellent reviews for my work and communication skills
• I developed good problem-solving skills while working on the myriad of real-life problems my clients handed out to me. This helped me develop a strong level of confidence. I quickly learned whatever technology or tool was required for the task.
• Fiverr profile: https://www.fiverr.com/users/dev_droid/