Work
Timeline
-
Economics of Transformative AI Fellowship Facilitator
Facilitating the Autumn 2026 fellowship for University of Chicago undergraduates.
-
Minerva Mentorship Program Student mentor
Mentoring two students per semester; wrote a mentorship framework for students joining startups.
-
University of Chicago MS in Financial Mathematics
Concentrations in financial computing (HPC, C++) and machine learning.
-
Intercontinental Exchange C++ Developer Intern
Wrote a new C++17 data-processing architecture for volatility surface construction.
Led AI best-practices documentation for the Curve Engine C++ team.
-
7 Chord Quantitative Researcher, UChicago Project Lab
Led a greenfield pipeline collecting 100+ US Treasury data fields into MongoDB and PostgreSQL.
-
Manteio Capital Quantitative Researcher, UChicago Project Lab
Built a PyTorch signal model for return forecasting and led portfolio implementation in a team of six.
-
Eve Air Mobility Sales Engineer, internal tools and operations modeling
Cut peak memory by 90% and processing time by 75% in a SageMaker toolchain.
Monte Carlo and sensitivity analyses behind $350M+ in signed sales agreements.
-
Alignment Research Engineering Accelerator (ARENA 3.0) Study group coordinator
Ran weekly Bay Area meetups on mechanistic interpretability; replicated "Indirect Object Identification in GPT-2 Small".
-
Develop for Good Data Engineer (volunteer)
Led a database migration and Tableau dashboards evaluating programs that reach 10,000 beneficiaries.
-
Eve Air Mobility Data Science Intern
Built named-entity recognition models that halved routine competitor-analysis time.
-
Minerva University Computational Sciences Curriculum Developer
Applied science-of-learning principles to statistics and algorithms courses; co-wrote interactive Jupyter notebooks.
-
Minerva University BS in Computational Sciences
Capstone on named-entity recognition for corporate news; led PyTorch NLP and deep RL tutorial sessions.
Projects
-
fast-options-pricer
A header-only C++20 library that prices European options under Black–Scholes three ways, in closed form, by Monte Carlo and by Crank–Nicolson finite differences.
Hard part: Making the slow methods fast and accurate. Antithetic and control variates cut Monte Carlo variance roughly 20–100×, and a Rannacher-started, pre-factored Crank–Nicolson solver prices an 800×200 grid in about 0.6 ms.
-
fbm-fast-pricers
A Monte Carlo pricer for arithmetic Asian calls under rough fractional stochastic volatility, comparing three ways to sample fractional Brownian motion.
Hard part: Path generation dominates the cost. Circulant embedding with FFT overtakes dense Cholesky at about 250 time steps and is 7.6× faster at 4,000; a low-rank rSVD sampler is faster still but underprices by about 4%.
-
stratum
A three-layer, Hyrule-inspired map for deep technical study, starting with HPC. You clear shrines by committing a small build and a write-up, and the map is rendered from the repo.
Hard part: Keeping the repository as the only source of truth, with no database or browser storage, while rendering a 217-entry, 22-region world as an explorable map.
-
orderbook-reconstructor
Rebuilds a 10-level market-by-price order book from a Databento market-by-order feed, with a parser, a reconstruction engine and a writer.
Hard part: Parsing a high-volume feed without allocating, using string_view fields over a fixed 256-byte line buffer sized from the maximum field lengths.
-
the_hack_computer
The nand2tetris computer built from logic gates up, covering the ALU, memory, CPU and a working computer, plus an assembler written in Python.
Hard part: Building every layer only from the one below it, from NAND gates through the CPU to an assembler for its instruction set.