The early years.
Every generalist has a stretch of years where the throughline only appears in hindsight. Mine runs from private equity case studies to animated math videos to an AI document-querying system to three decades of water-access econometrics — and the throughline turned out to be: pick an unfamiliar domain, get functional fast, ship something measurable.
national surveys analysed across 3 decades of water-access data
answer retrieval accuracy on scanned-form document AI system
forms processed daily, team of 15, 85% accuracy
IIM Bangalore · Water-access research
Drinking-water access in India across 1992–2020 — three decades of survey data, one question: who actually gets water, and does government policy change that?
First stint (Dec 2023 – Jan 2024) · Research Intern:
- Organized and analysed 15 national surveys (NFHS, NSSO, IHDS) spanning 1992–2020, finding a 17% national increase in tap-water access — with 64% of rural households still without access as of 2020.
- Exposed the gender disparity in water fetching: 75% of the burden borne by women, with male involvement rising from 10.5% to 21% over the period — a positive but still-insufficient trend.
Second stint (Jan 2024 – Feb 2024) · Research Assistant:
- Ran heterogeneity analysis to identify variations in water access across states, castes, and income groups — finding systematic patterns that simple national averages masked.
- Revealed that government scheme impact was negligible: <1% of households receiving help, and that summer water shortages from sources remained pervasive regardless of official scheme coverage.
- Identified target states for causal policy analysis and assessed the feasibility of integrating IMIS administrative data for future research on policy effectiveness.
Aqua Vision National Testing Labs · Summer Intern
An environmental testing lab where I wore two hats: building an AI document system from scratch, and managing the team running the manual data pipeline it was meant to replace.
- Built an extractive document-querying system using Haystack and FarmReader — scanning 10+ form types, pre-processing with Haystack's Pre-processor pipeline, and retrieving the top-5 most relevant answers to free-text queries at 75% accuracy.
- Managed a team of 15 employees handling daily entry of 2,000+ forms, maintaining up to 85% accuracy — doing root-cause analysis on recurring data problems and implementing corrective processes.
- Recognized for exemplary performance with a Director-level Letter of Recommendation.
Vizuara · Mathematics Content Developer
Created concept-explanation videos in Manim — the same Python library Grant Sanderson uses for 3Blue1Brown — for a US ed-tech company building 3D, gamified, AR/VR-powered math content. The goal: make students think computationally, not just procedurally.
Finlatics · Research Intern
An IB / PE / VC-oriented research program — my first structured exposure to how capital actually moves through markets.
- Authored case-based projects on successful private equity exits, analysing M&A transactions in Indian and international markets — deal structure, entry/exit multiples, and value-creation levers.
- Studied the stages of PE investing from a fund's perspective: deal sourcing, due diligence, portfolio management, and exit strategies.
- Assessed investment proposals and screened startups for high-growth potential — building the analytical instinct for what makes a business fundable.