The Pennsylvania State University
Expected May 2028
B.S. Computer Science, Minor in Physics, AI · Schreyer Honors College
University Park, PA
- GPA: 3.71 / 4.00
Résumé
Updated August 2026 · Download the PDF
Education
Research
Independent · PyTorch · LessWrong
- Explained the one architecture Linsefors & Bushnaq left unresolved in their memorization challenge: a training-dynamics failure, not an expressivity limit — a fact whose hidden units all die has its gradient gated to exactly zero
- Split their conflated bias flag for the discriminating experiment — an out-projection bias rescues completely (1024 facts), an in-projection bias not at all (488) — overturning the published account; a transplanted 672-fact witness rules out expressivity outright
Independent · PyTorch, TransformerLens · LessWrong
- Disconfirmed two of my own hypotheses — quotation and parenthesis grammar — with targeted QK attention-pattern controls
- Trained a 6-block character-level transformer and isolated a copying head via OV-circuit eigenvalues and copying scores
- Found the OV circuit encodes bracket completion while the QK circuit stays blind to it: a half-built mechanism
Independent · TransformerLens
- Recovered the IOI circuit (Wang et al.) with direct logit attribution, activation patching, and path patching, each hand-implemented
Monte Carlo Radiative Transfer — Neutron-Star Pulse Profiles
Oct 2025 – Present
Penn State Multi-Campus REU · advised by Prof. Asif ud-Doula · Python
- Built a Monte Carlo scattering atmosphere feeding a general-relativistic pulse-profile model (Schwarzschild light-bending, Doppler) to quantify the systematic from the isotropic-emission assumption in neutron-star mass–radius measurements
- Validated against Chandrasekhar's exact H-function (reduced χ² = 0.70) and Bogdanov et al. waveforms (0.11% peak error)
- Showed the systematic routes between pulsed fraction and waveform shape; first-author paper in preparation
Experience
Nittany AI Advance Internship Program
Dec 2025 – Present
Application Specialist · Client: Penn State Office of the Physical Plant
State College, PA
- Scoped ambiguous client requirements into technical documentation for the Docfinity system
- Prototyped evaluation protocols for generative and OCR-derived content; drove requirements to consensus ahead of launch
Penn State Department of Physics
Aug – Dec 2025
Physics Learning Assistant · PHYS 211
University Park, PA
- Taught calculus-based mechanics in problem sessions, labs, and exam prep
Projects
CrashAI — Crash-Risk Analysis Platform
Feb – Jul 2026
Founder & Technical Lead · Scoped with PennDOT safety staff · Python, TypeScript
- Built a crash-severity model — XGBoost over 81 roadway and behavioral features — with per-site SHAP attribution and what-if simulation
- Led a team across the stack: a 33-tool agent reasoning loop, RAG over agency engineering manuals, a hash-verified artifact pipeline, and an eval harness
Skills
ML & Research: Python, PyTorch, HuggingFace Transformers, TransformerLens, NumPy, Scikit-learn, Git
Interpretability: Transformer circuits, induction heads, OV/QK analysis, activation and path patching, direct logit attribution
Methods & Tools: Monte Carlo simulation, numerical validation, linear algebra · Java, JavaScript / React, SQL, Flask