Portfolio
Endeavors
I have worked in three environments that look unrelated from the outside: AI research, semiconductor process engineering, and quantitative research. The common thread is measurement under uncertainty: building tools, experiments, and analyses that make complicated systems easier to reason about.
I am still adding hard numbers where I can share them publicly. For now, this page focuses on what I personally built and the shape of the problems I worked on.
AI Evaluation and Reasoning Infrastructure
Axiom Math | Research Engineer Intern to Member of Technical Staff | 2025-2026
I designed and ran evaluations for mathematical reasoning models, then worked on the PyTorch-based infrastructure that supported those experiments at larger scale. The work sat between research and tooling: benchmark the models, identify the failure modes, and make the evaluation loop reliable enough that other researchers could trust what it was showing them.
The hard part was avoiding fake clarity. A math model can arrive at the right answer for the wrong reason, fail for several different reasons at once, or look strong on a benchmark that is too narrow. I spent a lot of time thinking about problem classes, alternate solution paths, and how formal verification could make evaluations more useful than a simple score.
This role shaped how I think about building: the best tools do not just automate work, they make a system more legible.
Semiconductor Process Engineering
Lam Research | Process Engineer I, Epitaxial Sustaining R&D | 2024-2025
At Lam, I worked on epitaxial deposition processes where small parameter shifts can become real manufacturing problems. I used design-of-experiments methods and statistical analysis on process data to investigate wafer-to-wafer and lot-to-lot variability.
Root cause work in a fab environment is messy in a useful way. The answer is rarely one variable. It lives somewhere between the tool, the recipe, the wafer history, the sensor data, and the people who understand the equipment. I worked with equipment, integration, and process engineering teams to diagnose excursions and turn findings into changes that could survive production use.
Some of that work was adopted into standard operating procedures for tool qualification and sustaining readiness. More importantly, it taught me how different real systems feel when they push back against theory.
Quantitative Research and Model Analysis
Greywolf Capital Management LP | Research Intern | Summer 2024
At Greywolf, I worked with noisy financial and model data from two angles: order book tick data and high-dimensional neural activations. I engineered market microstructure features, used t-SNE to look for latent structure in activations, and built Sim2Real domain adaptation pipelines with morphological preprocessing.
The challenge was not finding patterns; it was deciding which patterns deserved belief. In markets especially, visualization can seduce you into overreading noise. I learned to treat plots as instruments for asking sharper questions, not as evidence by themselves.
The work fed into probabilistic model analysis and trading strategy research, and it gave me a lasting respect for data that refuses to sit still.