# Parker Patton

574-371-6442 | jppatton@crimson.ua.edu | https://www.linkedin.com/in/jparkerpatton | https://www.ycty.ai

Canonical PDF: https://www.ycty.ai/Parker_Patton-Resume.pdf

## Education

**The University of Alabama** — BS/MS Pure Mathematics, August 2025–present. Expected graduation: December 2028. Tuscaloosa, Alabama.

Fall 2026 coursework: Real Analysis, Data Structures & Algorithms, Calculus III, Differential Equations.

## Experience

### LLM Interpretability Research — Algoverse AI Research

May 2026–present · Remote

- Investigating unverbalized evaluation awareness in LLMs under Madhur Panwar, targeting an ICLR submission.
- Developed an auditable dataset pipeline yielding 2,880 bilingual English/Japanese prompts; implemented Qwen and Gemini cross-model generation, blind verification, and deterministic semantic and formatting QC.
- Designed a six-level incentive titration across bilingual scenarios to separate explicit evaluation recognition from latent behavioral adaptation.

### AI Safety Lab — The University of Alabama

September 2025–present · Tuscaloosa, Alabama

- Developed a low-dimensional latent-space attack using video-to-video models for downstream classifiers; executed 3,652 trials across 686 source videos and evaluated every result after 8-bit GIF serialization.
- Achieved 89.5% ASR (265/296 eligible Kinetics-400 clips) at 27–30 dB PSNR; mapped the dimension-fidelity frontier, with 47% ASR from 16 randomly sampled latent dimensions.
- Characterized robustness across perturbation bases, video VAEs, and classifiers; 44/90 successful attacks remained misclassified after CogVideoX re-encoding.

### Financial ML Research — The University of Alabama at Birmingham

June 2024–May 2025 · Birmingham, Alabama

- Developed a volatility-conditioned DDPM (PyTorch) to generate plausible price paths across eight assets.
- Found higher point-forecast MAE than ARIMA (0.083 vs. 0.066), while demonstrating superior uncertainty calibration (CRPS 0.057; Diebold-Mariano p = 0.002) and realistic variance structure.
- Presented research at ASFA Symposium; awarded Honorable Mention in CARSEF Math & CS.

## Projects

### CS2Desk

Python, TypeScript, Next.js, SQLite · July 2026–present · https://github.com/catears124/cs2desk

- Built a CS2 trade-up search engine that ingested 147,000+ Steam listings and 2.25M bid levels, evaluated 151,000+ contracts, and ranked opportunities by expected profit, risk, and market depth.
- Modeled Steam fees, float-constrained outcomes, bid-side liquidity, and fillability to reject nominally profitable but non-executable contracts.
- Deployed a Python/SQLite optimizer with a Next.js dashboard; searches 50,000 contracts in under two seconds.

### osu!rankguess

Python, PyTorch, FastAPI, ONNX, PostgreSQL · March 2026–present · https://osurankguess.com

- Built and published osu!3k, a multimodal dataset of 3,000 gameplay videos, replay files, and metadata across 1,767 players; automated replay ingestion, rendering, metadata joins, and quality filtering.
- Trained a CatBoost/PyTorch ensemble achieving 0.846 R² and 1.84-point MAE on an unseen test set.
- Deployed a FastAPI/ONNX platform on Vercel with replay parsing, CPU inference, and PostgreSQL storage.

### Independent Smart-Contract Security Research

Solidity, Foundry, EIP-712, HyperEVM · June 2026

- Responsibly disclosed two settlement vulnerabilities in live options contracts by reverse-engineering EIP-712 signatures, packed calldata, and balance-debit paths.
- Found 38 post-expiry settlements across 800 live transactions and reproduced both flaws on a mainnet fork.

## Technical Skills

- Languages: Python, C/C++, TypeScript, SQL (Postgres, SQLite), Java
- Frameworks: PyTorch, NumPy, pandas, Next.js
- Quantitative: Diffusion Models, Time-Series Forecasting, Statistical Testing, Financial Data Engineering
- Developer Tools: Git, Docker, Google Cloud Platform, AWS, Linux, Vercel, Supabase, GitHub Actions