Jan 2022 – Jan 2024 · 2 yrs

• Built a suite of products with the unifying theme of extracting "insights" from bank data. Got one product from 1->10, another from -1 -> 1, and the third from 10 -> 10 (yes).

Jan 2019 – Jan 2022 · 3 yrs · Mountain View, California, United States

• Acquired by Snowflake.
• 2021-22: Led an 8-person cross-functional team to bring Neeva to the workplace. Wore multiple hats—management, product, data science, and engineering—from time-to-time to get the job done.
• 2019-21: Joined the team as a founding engineer and built many of Neeva's differentiating search product features from the ground up, while contributing across the entire stack (from dev-ops all the way to front-end).

Jan 2017 – Jan 2019 · 2 yrs · Mountain View, California, United States

• 2019: Launched the super chat attribution product in YouTube Live videos, as part of the alternative monetization team.
• 2017-19: Worked on several applied ML projects to 2X the top-of-the-funnel activation rate for interactions with the third-party platform in Google Assistant.

Jan 2016 – Jan 2016 · 0 mo · Mountain View, California, United States

• We came up with a flywheel algorithm to automatically grow our proprietary ontology of product categories—no human intervention required! The size grew 4X during the time of my internship.
• Stepped up as the product manager for my second project where we built a new data viz tool for use in QA. We conducted usability studies and demonstrated an increased e2e efficiency of 20%.

Jan 2014 – Jan 2014 · 0 mo

• Improved the inference efficacy of Microsoft's internal tool R2, by 10X in comparison to benchmarks such as Church, Venture and Infer.net. The leap came from adopting Hamiltonian MC in conjunction with a few tricks.

Jan 2013 – Jan 2014 · 1 yr

• Spent 2 years struggling with the synthesis modulo bisimulation problem which didn't go anywhere—had to quit to keep my hairline intact

Jan 2013 – Jan 2013 · 0 mo

• We improved the tightness bounds of the VC dimension in the art gallery problem under a symmetric setting. Took us an entire bundle of paper to get there.

Jan 2012 – Jan 2012 · 0 mo

• Took an unfruitful crack at improving the upper bound of block sorting algorithm, which has applications in genomic rearrangements and bridge (one of which I was very interested in).