Oct 2021 – Dec 2024 · 3 yrs 2 mos · Bengaluru, Karnataka, India · On-site

• Head of applied sciences + AI applications at Jupiter. Honestly, one of the most exciting and fulfilling rides of my career.
• Super lucky to have amazing peers and mentors, and to be part of Jupiter’s early days. Built and led this incredible mix of scientists, engineers, annotators — all obsessed with one thing: creating the best personal finance experience out there. Along the way, learned a ton about data, ML-driven products, and how to build a culture that puts customers first. We shipped some wild stuff — SOTA merchant intelligence models for India, custom SLMs that were cheap but insanely accurate, models that dug gold out of messy alternative data. Later we fed into risk + user-growth + recsys models, LLM-based agents for automated insights, even recurrent transactions before they happened (think subscriptions, income, and user behavior) -- impact was huge — PFM scaled to 0.XM MAUs with crazy retention, credit approvals up, losses down, increased targetable base by XX. Felt like we were building magic, and doing it together.

Sep 2014 – Sep 2021 · 7 yrs · Bengaluru, Karnataka, India

• co-founder at Daisho. Started with the dream of making ML accessible to anyone. Daisho made machine learning push-button simple. upload your data, answer a couple of questions, and boom — insights, models, predictions, all done automatically. no data science team, no coding, no fuss. just raw data turned into decisions in hours, not months.
• Built a crazy fast algo layer in python + C (20-50X+ speedup over existing solutions) capable of solving any problem in any domain (autoML, before what it's now) - zero dependence on sklearn or external packages (except for the FORTRAN linalg part!). This is 2013ish .. existing packages either wouldn't scale to 100s of Gigs of data or would be notoriously slow -- and we were building for users with the least patience -- the business users.
• Scaled as much as we could, using a build-operate-transfer platform play with some of India's largest companies (think X$B annual revenue cos in India). Wore every hat along the way — sales, tech, customer success. pure 0→1 grind, huge learning curve, and an exit at the end. Currently it's driving single digit $M revenue growth in acquired entity.

Jul 2013 – Sep 2021 · 8 yrs 2 mos · Bengaluru Area, India

• Pretty much the same as Daisho .. but the SaaS version that we tried for business analysts first.

Aug 2012 – Jun 2013 · 10 mos · Bengaluru Area, India

• Responsible for the unstructured data / text mining platform within I&D. Mostly built a ton of prototypes on top of ML models which we built. Dabbled in supporting pre-sales pitches for enterprises.

Jan 2008 – Dec 2011 · 3 yrs 11 mos

• Lead Collaborator at Virginia Tech’s Laboratory for Interdisciplinary Statistical Analysis (LISA) — consulted on 150+ projects, taught short courses on advanced methods (high-dimensional data analysis, random effects). Research in Bayesian model selection, efficient MCMC, high-dimensional visualization, and pathway data modeling.

Feb 2005 – Dec 2007 · 2 yrs 10 mos

• Modeling, evaluating and monitoring of transaction level credit risk (member of a small team responsible for over $1B in transaction volume)
• Portfolio credit risk - Credit line management for the credit card portfolio of JP Morgan Chase
• Active interaction with other parts of risk management e.g. acquisition risk and marketing partners
• Exposure to risk scoring methodologies, catastrophic risks and emerging drivers of risk

Jan 2003 – Jan 2004 · 1 yr