UW–Madison M.S. candidate in Psychology: Data Science in Human Behavior (expected Dec 2026); B.S. in Psychology (neuroscience concentration, sociology minor), UW–Parkside. Applied AI practitioner building full-lifecycle systems spanning multi-agent document-intelligence pipelines, transformer fine-tuning and calibration, LLM evaluation harnesses, computer-vision classification, and behavioral prediction models. Program capstone with American Family Insurance as industry partner, developed through biweekly working sessions with their Enterprise Data Science lead. Five years of high-volume operations experience in quality assurance, documentation, and process improvement.
Education
University of Wisconsin–MadisonMadison, WIAugust 2025 – December 2026
M.S. Candidate – Psychology: Data Science in Human Behavior · GPA: 3.85
Graduate program in which roughly 95% of coursework is data science and data engineering — building on an undergraduate foundation in neuroscience research, experimental design, and statistical analysis.
Advanced training in machine learning (neural networks, regularized and ensemble models, transformer fine-tuning and LLM evaluation), statistical modeling, and large-scale data systems (Spark, gRPC services, HDFS, Parquet, SQL, Docker).
Industry capstone with American Family Insurance (see Projects).
University of Wisconsin–ParksideKenosha, WISeptember 2020 – May 2025
B.S., Psychology
Concentration: Neuroscience. GPA: 3.93
Coursework spanning cognitive and computational neuroscience, psychological theory, clinical psychology, psychopharmacology, psychometrics, psychological statistics, and early data science preparation — plus philosophy, neuroethics, and pre-law.
Minor in Sociology; Provost’s List (7x), Dean’s List (8x), Distinguished Achievement Award (2025), and Psychological Research Certification (300 lab hours).
Conducted cognitive neuroscience research under Dr. Melissa Gregg, designing and pilot-testing an experimental study on truth, deception, and cognitive workload (study code: github.com/Exios66/truth-deception-architecture).
Operated EEG systems and analyzed neural time-series data to quantify brain activity, contributing signal-processing and statistical insights to ongoing lab research on sleep quality and cognitive load.
Presented findings at the 2025 Midwestern Psychological Association conference in Chicago and at Research in the Rotunda.
Meijer Distribution — Logistics CoordinatorPleasant Prairie, WIJul 2020 – Aug 2025
Promoted from Retail Loader to Logistics Coordinator for reliability in quality assurance and data-driven process documentation across high-volume operations.
Analyzed quality-assurance data for 64 daily deliveries across a Midwest regional network, maintaining error rates below 1% through root-cause investigation and process refinement.
Partnered with warehouse and transportation teams to investigate quality problems and resolve logistics exceptions before shipment.
Projects
Digital-MailroomData Science in Human Behavior capstone — American Family Insurance (industry partner)
Capstone developed with American Family Insurance as industry partner: 5 Python packages addressing document-intelligence problems scoped by their Enterprise Data Science division — entity extraction, classification, and scalable legal-document workflows.
Implemented a unified LangGraph pipeline integrating legal-document routing, prompt experimentation, deterministic scoring, corpus analysis, and deployment workflows in a reproducible Docker environment.
Developed a ModernBERT document classifier with end-to-end training, evaluation, and fine-tuning pipelines — including data lineage tracking, probability calibration, and production routing logic.
Implemented PyTorch training pipelines with Transformers and Hugging Face datasets, SciPy-based evaluation, and ONNX export for FastAPI serving on Modal; maintained pytest coverage throughout.
Developed a Python toolkit that converts TIFF/PDF pages to 300-DPI images, runs OCR, and classifies documents into 16 RVL-CDIP categories with OpenRouter vision models.
Tracked 29 prompt versions with confusion matrices, misclassification analyses, and cost projections to guide prompt and model selection.
Built a behavioral prediction pipeline on 241 complete respondents from 252 matched survey records — Ridge, Elastic Net, k-NN, Random Forest, XGBoost, and MLP predicting communication-apprehension scores from demographic, employment, geographic, and transit features, with cross-validation and SHAP feature importance.
Validated the PRCA’s psychometric structure with factor analysis, then built LLM ‘digital twin’ personas via tiered embodiment prompts (Ollama/OpenRouter/vLLM) and benchmarked them against the ML baselines on shared metrics (MAE, band accuracy, F1) — analyzing systematic bias by demographic group.
Neuron AtlasMechanistic interpretability · live browser model
Built a browser interpretability tool that maps all 8,192 MLP neurons of TinyStories-8M and supports live steering, ablation and clamping beside random-direction controls; recorded Qwen studies extend it beyond live browser inference.
Implemented a TypeScript GPT-Neo forward pass with a KV cache, a Three.js atlas and a Python sweep/analysis pipeline, verified against PyTorch and covered by Playwright tests for navigation, replay, phones and failed downloads.
MapReduce-ReloadedDistributed systems · Go + Python
From-scratch Go re-implementation of Google’s MapReduce (Dean & Ghemawat, OSDI 2004), extended for modern infrastructure: gRPC coordinator/worker protocol, Docker Compose cluster deployment, and a Python reference implementation sharing the same protobuf contract and file format.