I grew up fascinated by how systems work, which led me to pursue a B.Tech in Engineering at IIT Madras, one of India's premier technical institutions. There, I built a deep foundation in mathematics, statistics, and algorithms, and developed a taste for turning complex, messy data into decisions that actually matter.
During my undergrad, I joined Seat of Joy, a child safety startup incubated at IIT Madras, as a Business & Strategy Analyst. I built a probabilistic market-sizing model from Indian Census data (100+ tables, 200K+ rows each) that estimated 55M target customers with 5% YoY growth, developed a supply-chain optimization model using operations research principles, led full competitor and pricing analysis across the category, and represented the startup at Shark Tank India Auditions, pitching data-backed market and business strategy to investors.
I'm now pursuing my Master of Science in Business Analytics (MSBA) at UCLA Anderson, deepening my expertise in machine learning, data engineering, and optimization. I'm drawn to problems where rigorous analysis drives real-world impact, from production agentic RAG systems to large-scale data pipelines to deep reinforcement learning.
Becton Dickinson
Seat of Joy (Incubated at IIT Madras)
Child safety startup developing a full-body protective seat for two-wheelers, addressing the 2 children lost daily in India to two-wheeler accidents.
India · During Undergrad
A selection of data science and ML engineering work.
A production ML pipeline that forecasts theme park wait times a week ahead, retrained unattended every week on Databricks and served live for $0/month. Powers the Mapblazer routing engine.
MotivationA route built on 9 AM wait times is already wrong by 1 PM, because crowds shift all day and one bad estimate cascades into a ruined plan. I built the forecasting system that predicts wait times across the whole week, so Mapblazer's routing engine can anticipate crowd flow instead of chasing it.
An exact Gurobi MILP that plans a whole park day at once: which rides to do, in what order, and at what time, with every queue priced by the 30-minute slot you actually arrive in. The routing engine behind Mapblazer.
MotivationA park day is not a shortest-path problem: a ride costs whatever its queue costs, the queue changes hour by hour, and the order you pick changes the costs that decide the order. I wanted to solve that feedback loop exactly rather than greedily, with an objective that says when idling for a cheaper slot is worth the wait it costs.
A ViT-style Deep Q-Network that builds a month of legal airline crew pairings one leg at a time, feasible by construction, published as an interactive replay you can scrub decision by decision.
MotivationCrew cost is second only to fuel for most airlines, and the classical column-generation approach has to be rebuilt for every schedule. I wanted to know whether a learned construction policy could obey a real rulebook exactly, generalize to a month it had never seen, and trade a little coverage for solutions that don't shatter when one flight runs late.
A regression discontinuity re-test of whether Airbnb's Superhost badge still moves host outcomes now that Guest Favorite has taken over discovery. Five of six outcomes show no jump at the 4.8 cutoff; only review volume does.
MotivationPublished research from 2023 found the Superhost badge causally lifted bookings and revenue at the 4.8-rating cutoff. Then Airbnb launched Guest Favorite and removed the Superhost search filter, so the discovery mechanism behind that finding was gone. I wanted to run the same design on current data and report the answer it gave, not the one that would make a nicer story.
Production-deployed agentic RAG system for UCLA MSBA students to query course materials (lecture slides, transcripts, and PDFs) using natural language. Live at tirth-courserag.duckdns.org.
MotivationThe UCLA MSBA program runs 4 simultaneous courses, each with its own slides, transcripts, homework deadlines, and deliverables spread across a shared Google Drive. Students constantly lose time hunting for information manually. I built a fully agentic system that classifies every query, self-verifies deadline answers, supports human-approved file uploads, and can explain exactly which source chunks drove any answer, deployed at effectively zero infrastructure cost on Oracle Cloud Free Tier.
End-to-end data engineering pipeline correlating weather patterns with Yelp restaurant sentiment using PySpark, Snowflake, Airflow, and Tableau, processing 2M+ records.
MotivationCurious whether weather drives restaurant ratings and business patterns, I built a production-grade data pipeline ingesting the full Yelp Academic Dataset and OpenWeatherMap API, performing distributed ETL at scale, NLP sentiment scoring, and surfacing insights through an executive Tableau dashboard.
Deep reinforcement learning agent that solves the Maltese Gear Cube using a CUDA-accelerated neural heuristic.
MotivationTo apply the DeepCubeA algorithm to a novel, higher-complexity puzzle and validate whether deep RL can generalize to unseen combinatorial state spaces.
The tools and technologies I work with.
Whether you're recruiting, collaborating, or just want to talk data, my inbox is always open.