Feb 2022 – Feb 2023 · 1 yr

• After realizing humit wasn’t working, we pivoted. We were toying with NFTs as a way to represent exclusivity, fandom, and access—potentially unlocking a new revenue stream for artists amid declining streaming income. We experimented with creating our own music IP, granting ownership rights, allowing fans to manipulate stems to craft unique versions, and enabling direct interactions with artists whose NFTs they held. The entire market got wiped out from the crypto crash. Also made us realise that most of crypto, at least, then was about speculation and nothing more.

Jan 2021 – Apr 2022 · 1 yr 3 mos · Bangalore Urban, Karnataka, India

• At humit, we set out to create a fun, music-centric social experience—think Reddit for music—bringing together users from different streaming platforms under one roof. A classic ZIRP-era idea that wouldn’t get funded today, but a formative journey nonetheless. We learned what it takes to ship great software and became proud sticklers for delightful UX and pixel perfection. We also had a bunch of really passionate users.
• Backed by Antler, FirstCheque, and angels including Kunal Shah
• Acqui-hired in 2023 by Juno Finance (backed by Sequoia & Dragonfly Capital)

Feb 2020 – Jan 2021 · 11 mos · Mumbai Area, India

• Worked on PayU's BNPL product LazyPay mainly focussing on improving the risk models being used to underwrite millions of customers.
• Here are some of the things I worked on :
• Graph based approaches to risk modelling using network data. This helped us almost 2x the customer base for credit approval.
• Designed and implemented the inference pipeline for LazyPay’s risk engine that underwrites hundreds of millions of customers every month.
• Built tools to better analyse the tons of models we were training.

May 2018 – Dec 2019 · 1 yr 7 mos · Mumbai, Maharashtra, India

• Worked on implementing deep learning solutions for Fortune 500 companies, primarily focusing on NLP applications. Much of what we see Gen AI doing today—Q&A, text summarization, and more—felt further away from reach at the time. This was before the era of scaling laws, where most of the models claiming to do them were toy repositories with cherry-picked examples. And making anything work, required crazy amounts of data to even have a chance.