Tanmay Kuchhal · University of Washington
I do AI research, build with data, and think a lot about markets, products, and access. Also: tennis, chess, and an unreasonable number of playlists.
01 · Introduction
I study Informatics and Economics at the University of Washington. Two lenses, one question: how do systems shape what people can actually do?
Right now my work centers on AI research. At the UW Information School I benchmark the reliability of large language models with Prof. Ott Toomet, and at the UW Math AI Lab I'm building LeanCoT, a framework that formally verifies LLM chain-of-thought reasoning in Lean 4. Before that, I built regression and gradient-boosting models on Formula 1 data and designed the architecture for a supply-chain traceability platform.
Away from the screen: tennis, chess, playlists on repeat, and a long-running fascination with the interwar years. I grew up between New Delhi and Dehradun before landing in Seattle, and I like to think all three cities show up in how I build.
Off the clock
02 · Foundations
Born in New Delhi, India.
2023 — 2027
B.S. Informatics + Economics · Class of 2027 · Seattle, Washington, USA
2017 — 2023
International Baccalaureate Diploma · Dehradun, Uttarakhand, India
Kashmir House
2012 — 2017
Noida, Uttar Pradesh, India
2009 — 2012
Dwarka, New Delhi, India
2007 — 2009
Dwarka, New Delhi, India
03 · Selected Work
Case studies across machine learning, systems design, and fintech product thinking.
Data Science · Regression Analysis
Predicting fastest lap times in Formula 1 from 5+ years of multi-season race data, comparing linear, interaction-effects, and gradient-boosting models.
Product · Systems Design
A systems-design project for end-to-end transparency in fashion supply chains — tracing garments from raw material to retail using blockchain, IoT, and cloud architecture.
AI Research · Formal Verification
A neuro-symbolic framework formalizing LLM Chain-of-Thought reasoning in Lean 4 — catching logical errors that First-Order Logic solvers like Z3 structurally cannot express.
I like problems where the technical answer and the human answer have to agree.
04 · Toolkit
What I bring across product, data, and business.
05 · Path
Jun 2026 — Present
Building LeanCoT, a neuro-symbolic framework that formalizes LLM Chain-of-Thought reasoning in Lean 4 to catch logical errors beyond the reach of First-Order Logic solvers like Z3. Developing a constrained domain-specific language in Lean 4 that maps natural-language reasoning steps to formally verifiable proof states, so semantic errors become direct feedback signals for correcting LLM reasoning traces.
Apr 2026 — Present
Building Python evaluation pipelines to benchmark prompting strategies — Chain-of-Verification, few-shot, zero-shot — across three open-source LLMs on contextual QA datasets, tracking 10+ performance metrics. Aggregated and quality-controlled thousands of model outputs with reproducible, version-controlled pipelines; results contribute to ongoing iSchool research on LLM reliability.
Jul — Aug 2024
Integrated geospatial and environmental datasets to monitor sustainable development projects across 4+ sites, contributing to a data platform serving field teams and external partners. Structured raw environmental data into actionable reporting dashboards.
06 · Beyond the Classroom
President · The Doon School
Panchayat Ghar is The Doon School's community outreach centre, where students teach children from nearby communities and the families of school support staff. As President, I led its programming: career guidance sessions, aptitude testing, and specialized classes aimed at opening real pathways to higher education and vocational training.
07 · In Print
Peer-Reviewed Paper
Journal of Social Science and Economic Research
Applied chi-squared tests on OECD trade datasets to quantify the impact of bilateral geopolitical decisions on trade flows.
Read the paperBook
Notion Press
Featured in Writing Geeks Magazine's 'Top 10 Books'; 200+ copies sold across Amazon and Flipkart.
08 · Get in Touch
Let's build something worth shipping.
Open to internships, product roles, research collaborations, and interesting conversations about tech, fintech, and building things that matter.
The fastest way to reach me is a direct email — I read everything.
Email Me(Site owner: add your Formspree ID in content/site.ts to enable the contact form.)