Mar 2020 – Oct 2020 · 7 mos

• Took the Tapestry project from conception to deployment during the peak of COVID-19 within only a few months. Developed a method for fast, low-cost pooled RT-PCR testing by combining Combinatorial Group Testing and Compressed Sensing, and an app for easy manual pooled testing. Collaborated with a diverse team of computer scientists, engineers, and biologists. Mentored junior researchers and engineers on various aspects of the project. Tapestry was a finalist in the X-Prize Rapid COVID Testing Challenge 2020, Open Innovation Track.

Jul 2017 – Aug 2024 · 7 yrs 1 mo · Mumbai · On-site

• My thesis focused on different aspects of Compressed Sensing and Group Testing. Derived Tapestry, a new algorithm for pooled COVID-19 RT-PCR testing which combines both combinatorial group testing and compressed sensing, proved its theoretical properties and implemented it on the field during the peak of the pandemic. This led to the formation of a molecular diagnostic startup called Algorithmic Bioloigics, and Tapestry was also a finalist at the X-Prize Rapid COVID testing competition 2020, Open innovation track. Created a Quantitative Matrix-Pooled Neural Network (QMPNN) for Efficient Deep Learning (classification and outlier detection) using group testing and compressed sensing. Derived a cross-validation based algorithm for joint recovery of a graph and a graph signal from compressive measurements, using which images with edges can be recovered more cleanly.
• Also involved in a number of other projects - high-level rule-exemplar supervision for deep learning, sparse adversarial example generation, endless novel pattern generation.
• Publications at IEEE OJSP, ICLR, EMNLP.

Jun 2016 – Jul 2017 · 1 yr 1 mo · Mumbai Metropolitan Region

• Implemented an Open-Ended Evolution (OEE) system (Decision Tree Organisms) to endlessly create novel 1-d patterns. Derived a replicator equation based model for learning and sampling from probabilistic graphical models. Posed a new metric for sophistication of patterns based on Algorithmic Information Theory (Kolmogorov Complexity).

May 2015 – Dec 2015 · 7 mos · Mumbai Metropolitan Region

• Took Shop101 from conception to launch, leading the backend engineering team. Designed the API server, ensuring data consistency, scalability and performance. Conceived and implemented key security features such as secure password storage and id obfuscation. Implemented features and fixed bugs for the Android app without prior Android development experience, to ensure timely product launch.
• Languages/Platforms: Java, SQL (PostgreSQL), Android, Linux

Jul 2011 – Apr 2014 · 2 yrs 9 mos

• Performance analysis of key components of the Whitewater (now NetApp SteelStore) storage appliance: NFS backup, Cloud Replication and Peer Replication. Attained near-theoretical performance (4-10x) after fixing all bottlenecks. Implemented key features: Peer Replication and Openstack backend support, automatic replication thread tuning and ETA estimation. Peer replication involved careful design to ensure data consistency on peer and cloud, while maintaining parallel replication for maximum speed. Miscellaneous debugging and codebase improvements: lock order violation checking, code coverage, unit test parallelization, test machine monitoring via nagios.
• Languages/Platforms: C++, Boost, POSIX, pthreads, FUSE, NFS