I'm a student at UC Berkeley working at the intersection of machine learning, analytics, and investing. I build systems that make hard problems tractable — from self-supervised vision experiments and generative model pipelines to Bayesian cost simulations and churn investigations on SaaS businesses.
This site collects equity research, project write-ups, and analytical work. I'm drawn to problems that require both quantitative rigor and clear communication — where the goal is to be precise about what you know and honest about what you don't. Outside of research and code, I follow markets with a long-horizon, durable-moat lens.
Recent research
SoFi is a legitimate developing compounder at an early but identifiable inflection point. The central thesis: SoFi's current 18% operating margin grossly understates its 5-year potential. The Financial Services segment inflected to 51% contribution margins in Q4 2025 — from near-zero in 2023 — on an asset-light model. The Technology Platform is an overlooked embedded-finance infrastructure asset the market is missing entirely.
DoorDash is widely analyzed as a food delivery company. It is actually constructing a full-stack commerce operating system — spanning logistics, subscriptions, merchant data, and advertising — funded by delivery economics. The advertising business is the unpriced optionality the market is missing.
Selected builds
AI-assisted valuation workflows for public equities, with scenario modeling, fair value outputs, and live market context in one interface.
A research comparison of representation-learning objectives, testing whether JEPA-style prediction produces more useful embeddings than reconstruction.
A compact personalization pipeline for learning a photographer-specific editing style without keeping a heavyweight generation model in production.