The Pennsylvania State University
Expected May 2028
B.S. Computer Science, Minors in Physics and AI · Schreyer Honors College · GPA 3.71 / 4.00
University Park, PA
Résumé
Updated September 2026 · Download as PDF: research · engineering
Education
Research
Do Residual-Stream Probes Survive Degrading Chain-of-Thought Legibility?
Sep 2026 – Present
SPAR · mentored by Marios Tsatsos, with two fellow mentees · PyTorch
- Comparing three published readability scores for a model's written reasoning on one set of traces, to find whether they agree
- Testing whether classifiers on a model's internal activations still catch harmful reasoning as that writing degrades
Independent · PyTorch, jacobian-lens · Qwen3, Pythia, GPT-2, Gemma-2
- Traced the per-token bias in a lens that reads a model's internal activations as words to the model's own word frequency (Spearman 0.48); subtracting it made readouts 3.5–7× worse
- Replaced subtraction with z-scoring: recovered hidden secret words from fine-tuned models where the published method scored zero, and beat it 0.805 to 0.665 on Gemma-2-9B
Independent · PyTorch · LessWrong
- Explained why one small transformer in a published memorization challenge could barely learn: its ReLU units died in training and cut off the gradient, a training failure, not a capacity limit
- Confirmed it with one controlled experiment (a bias after the hidden layer restored full capacity, 1024 facts; before it, none, 488) and hand-built weights storing 672 facts; overturned the authors' explanation
Independent · PyTorch, TransformerLens · LessWrong
- Trained a 6-layer character-level transformer on Nietzsche's prose and located the one attention head that copies earlier text forward
- Ruled out two hypotheses about the head (quotation and parenthesis matching) with targeted controls
- Found the head had learned half of a bracket-matching mechanism: it knows what to write, but never learned where to look
Monte Carlo Radiative Transfer — Neutron-Star Pulse Profiles
Oct 2025 – Aug 2026
Penn State Multi-Campus REU · advised by Prof. Asif ud-Doula · Python
- Simulated X-ray photons scattering through a neutron star's atmosphere to measure the error a common assumption introduces
- Validated it against an exact analytic solution (reduced χ² = 0.70) and published waveforms (0.11% peak error)
- Showed the error shifts between two observables; first-author paper under review at Astrophysics and Space Science
Experience
Nittany AI Advance Internship Program
Dec 2025 – Present
Application Specialist · Team of 5 · Client: Penn State Office of the Physical Plant
State College, PA
- Built a Tesseract OCR + LLM + regex pipeline that extracts metadata (project name, ID, and more) from scanned project documents
- Raised extraction accuracy up to 10% by merging the LLM and regex paths and moving from GPT-4o to GPT-5 after benchmarking Azure's 4- and 5-series models; parallelized the pipeline for 5x faster runs
Projects
CrashAI — Crash-Risk Analysis Platform
Feb – Jul 2026
Founder & Technical Lead · Team of 3 · 1st of 40 teams, Nittany AI Challenge 2026 · Scoped with PennDOT
- Built a crash-severity model (XGBoost, 81 road and driver features) with per-site explanations and what-if simulation; led a team building a 33-tool agent, retrieval over engineering manuals, and an eval harness
Skills
ML & Research: Python, PyTorch, HuggingFace Transformers, TransformerLens, NumPy, scikit-learn, Monte Carlo simulation, Git
Interpretability: Transformer circuits, OV/QK analysis, activation and path patching, direct logit attribution, Jacobian and logit lenses