What I’ve learned, and how deeply.

Covered neural networks end-to-end — CNNs, RNN/LSTM, transformers, representation learning — with applied multi-modal projects. Our six-person team's final project, fusing LSTM time-series with FinBERT sentiment, came out of this class.

Classical ML (GLMs, tree-based methods, SVMs, ensembles) with heavy emphasis on experimental rigor — feature engineering, leakage prevention, honest validation. Earned A. The class's insistence on “just because you can use a transformer doesn't mean you should” was formative.

The shared foundation for every Berkeley data-science major. Covered the full stack: from EDA and SQL through regression and classification. The class's philosophy — “you should be able to explain why your model predicts what it predicts” — shaped how I build every model since.

Pre-MFE module on multivariate and stochastic calculus applied to derivative pricing. Heavy on manual derivations — re-derived Itô's lemma and Black-Scholes PDE from first principles. Gave me the math vocabulary to read any quant paper without getting stuck on notation.

Started with sigma-algebras and measure theory, built up through Brownian motion to martingale pricing and change of measure. The bridge between abstract probability and the P/Q measure distinction that matters for derivative pricing.

Cholesky, QR, SVD, and iterative solvers — with finance motivation throughout: covariance factorization, portfolio optimization, fast option Greeks. Implemented each algorithm to feel the stability and speed tradeoffs.

The prep course before Baruch's MFE program. Covered OOP in C++ applied to Monte Carlo pricing, binomial-tree and Black-Scholes implementations, and template design in a quant context. Distinction awarded for the final project; the course decided me on the Pre-MFE path.

Deep work on variance reduction, importance sampling, and copula-based risk modeling. Final project modeled tail risk under multiple correlated factors. The A+ was the result of spending twice the lecture hours in office hours debating method choices.

Covered linear, mixed-integer, and second-order cone programs. The duality chapter was where the abstract math finally met real decisions — why shadow prices matter, what LP relaxations actually tell you. Applied directly in the BTC ETF project's portfolio decomposition.

Taken concurrently with the ADR research project — the probability theory here directly grounded the statistical reasoning we used in the paper. First time I understood why a pricing model built on an SDE actually converges.

Covered relational algebra, query optimization, indexing, transactions and concurrency control. Project was a toy query engine implementation. This class is what made DuckDB feel like a reasonable tool to reach for in my BTC ETF project.

Earned A. Classic UCB data-structures class with the 3,000-line Java Gitlet project at its core. The project forced me to design a real object model before writing any code — that habit has stayed.