Oscar Zhu · St. Louis
I study Mathematical Sciences and Financial Engineering at Washington University in St. Louis. I build focused software that stays local, explains its assumptions, and is honest about its limits.
01 / Selected work
Quantitative finance · macOS 14+
An offline options laboratory that turns Black–Scholes–Merton formulas into a working, visual Mac app.
Price European options, inspect Greeks, solve implied volatility, shape multi-leg strategies, and stress spot and volatility together. The app stores work locally and does not require an account.
On-device translation · macOS 15+
A menu-bar translator for people who write between languages and want to keep the original sentence in view.
Select editable text, press Control–Option–T, and DualTyper inserts Apple’s on-device translation underneath. It rechecks the process, text, and selection range before editing and refuses secure fields.
The free build is ad-hoc signed and unnotarized, so each user must approve it in macOS Privacy & Security and grant Accessibility permission.
02 / ML reliability lab
01
Adversarial fixtures and high-precision references for numerical ML kernels. It makes a plausible-looking result answer to a stricter question: is it actually correct?
Explore releases02
An exact-oracle checker for affine INT8 quantization arithmetic, focused on round boundaries and cross-scale additions where small errors become model errors.
Explore releases03
A segmented monitor for approximate-nearest-neighbor indexes that looks for tail-query recall collapse instead of hiding it inside one average score.
Explore releases03 / About
I’m interested in tools that make complex work easier to inspect, not easier to hide.
At Washington University in St. Louis, I study Mathematical Sciences and Financial Engineering, with expected graduation in May 2029. That combination keeps me moving between models and interfaces: how something works, what it assumes, and whether another person can use it clearly.
My public projects are native Mac apps because I like software with a visible boundary. The code, data, and tradeoffs can stay close to the person using it.
04 / Principles
User data should stay on the device when the product does not need a server.
Secure fields, invalid inputs, stale selections, and model limits need explicit behavior.
Tests, release artifacts, checksums, and documented limitations matter more than a sweeping claim.
Good tools can be ambitious without becoming crowded.
05 / Connect