Machine Learning · Data Science · AI Engineering & Safety
I build AI systems from start to finish: multi-agent document pipelines, fine-tuned and calibrated transformers, LLM evaluation, and behavioral prediction models.
Madison, WI · Open to relocation
About me · Start here · Selected work · Background · How this repository is built
I'm finishing an M.S. in Psychology: Data Science in Human Behavior at UW–Madison (expected December 2026), after a B.S. in Psychology with a neuroscience concentration at UW–Parkside. Before graduate school I spent five years in high-volume logistics and operations, working on quality assurance, documentation, and process improvement. That is why I care about error rates, reproducibility, and cost as much as accuracy.
I'm interested in AI systems that know when they are unsure: calibrated classifiers, evaluation you can trust, and pipelines that pass hard cases to a stronger model or a person instead of guessing. Outside of that, I like turning physics into interactive software, and I'm curious about how people think and act, which is what my psychology training is about.
Each project has a case study on the portfolio.
- Digital-Mailroom — my capstone with American Family Insurance: five Python packages and one LangGraph pipeline for document routing, prompt experiments, scoring, and deployment.
- Mailroom-ML — a calibrated ModernBERT classifier that handles routine documents and knows when to hand off to an LLM.
- Wave Laboratory — a real-time 3D ocean and ship-motion sandbox that runs in the browser.
| Project | What it is | Stage |
|---|---|---|
| Digital-Mailroom | Five Python packages and one LangGraph pipeline for document routing, prompt experiments, scoring, and deployment | Capstone |
| Mailroom-ML | A calibrated ModernBERT classifier that handles routine documents and knows when to hand off to an LLM | — |
| Mailroom Reloaded | The mailroom rebuilt as one deployable service, with an audit log of every decision | — |
| RVL-CDIP-Classifier | Zero-shot document classification with vision LLMs, with tracked prompt versions and cost per page | — |
| Mailroom-Corpus-EDA | The data engineering behind the mailroom dataset: 3,302 documents from public sources, with provenance, label separation and an audit pipeline | — |
| AeroGlobe | A 3D flight-tracking globe built around a static data pipeline, with an early version live | In active development |
| MapReduce-Reloaded | A from-scratch rebuild of Google's MapReduce in Go and Python, with all 23 of the paper's features and chaos-tested fault tolerance | In active development |
| Psychometric Digital Twin | LLM "digital twin" personas against seven classical models on 241 survey respondents — the personas did not beat simple baselines | — |
| The Neural Architecture of Truth and Deception | A pilot study of nine people and 18 sessions with helpful and deceitful chatbot personas; the physiological, EEG and eye-tracking data are withheld until authenticated | — |
| MORNINGSTAR | A courtroom of LLM personas that argue, vote, and rule on engineering decisions | — |
| Wave Laboratory | A real-time 3D ocean and ship-motion sandbox that runs in the browser | — |
| Langton's Ant | A browser simulator for Langton's ant and other turmites on an unbounded grid, an early version with a pinned determinism test | — |
| The Mandalay Bay | A satirical casino resort simulator, play chips only | — |
- Education: M.S. candidate, Psychology: Data Science in Human Behavior, UW–Madison (expected Dec 2026). B.S. Psychology, UW–Parkside (2020–2025).
- Now: building services at UW–Madison Recreation & Wellbeing in Madison, alongside my capstone.
- Research: undergraduate researcher in a cognitive neuroscience lab at UW–Parkside (2024–2025), working with EEG and presenting at the 2025 Midwestern Psychological Association conference.
- Before that: logistics coordinator at Meijer Distribution (2020–2025), promoted from Retail Loader.
This repository is the Quarto source for the site — GitHub Pages serves the rendered pages straight from docs/, with no GitHub Actions. See how it's built for the page map, the render pipeline, and the design system.