Curriculum Vitae

Founder & Lead Developer. 50+ customer discovery interviews, NSF I-Corps Ideator (2x), Wond’ry Builder. $1,250 grant, $20K+ credits

React, Tauri, FastAPI, DuckDB, Polars, Ollama, XGBoost, PyTorch, ONNX Runtime

• Kept IRB- and FERPA-restricted data on the user's machine with a desktop app (Mac and Windows) that answers plain-English questions about spreadsheets; in use by a principal investigator at Vanderbilt and a small-business owner.
• Stopped the language model from inventing numbers: numeric questions run as exact DuckDB SQL, and an answer is marked verified only when exact code, not the model, produced it. Questions SQL cannot ground are declined, not guessed.
• Let users train, check and explain predictive models on their own spreadsheets without code: XGBoost search under a machine-sized time budget, PyTorch Lightning on DuckDB/Polars, checks for columns that leak the target, ONNX export.
• Prevented silent CPU fallback after finding builds that claimed GPU acceleration but ran on CPU: the inference backend detects NVIDIA, AMD, Intel Arc, Apple Silicon or CPU-only hardware and verifies which hardware the installed runtime really uses.
• Deterministically maps how Excel sheets connect (which columns and rows join): 0.91 precision, 0.76 recall on 2,950 files.

Independent Research Project, Meers Lab, Vanderbilt University Medical Center

Python, Evo 2 (7B genomic LM), GCP, GenBank/IGV/Geneious

• Won lab adoption and funding: researchers use the prototypes on their own sequences, and the principal investigator, Professor Chance Meers, is now funding development of this app as a Windows version (C# and WinUI 3 over the same Python pipeline).
• Cut the lab’s BLAST workflow from over 8 hours to under 5 minutes with two CLI prototypes that score every base of a DNA sequence with the Evo 2 genomic model and write Shannon entropy tracks for IGV and Geneious.
• Kept sequences and billing off my servers by running Evo 2 on a GPU in the user’s own Google Cloud project: one version creates and deletes the VM per run for cents, another keeps a box up for instant runs, or a local GPU if present.

Software Engineer, remote. Built the company’s calculator emulator and its competition suite

TypeScript, Tauri 2, Rust, Node.js, Vitest, Electron, Python, Flask

• Cut exam prep from 2 hours to a minute by automating the competition in a Flask + Electron app: Excel-driven 60-question exams and answer keys, an entry sheet whose dropdowns make typos impossible, percentile grading, student reports.
• Built a pixel-accurate emulator of the AB-991EX calculator for Windows, macOS and web: a framework-free TypeScript core whose parser and renderer drive the 192×63 display in 12 modes, decimal.js matched to hardware precision.
• Made any change that moves an LCD pixel fail CI with 244 screenshot-comparison files in a 3,560-test suite; built the Node telemetry backend with batched analytics, error ingest, per-IP and per-install rate limiting and the auto-update feed.

Lead Researcher, The University of Texas at Dallas, Richardson, TX

Python, Keras, XGBoost, MATLAB, Excel

• Led a 30-person research team investigating engineering identity in STEM students across data collection, classification, and predictive modeling; built the internal Python tooling and data-cleaning scripts, then trained and mentored the team on them.
• Saved the team 500+ hours by automating key data-pipeline segments in Python: keyword/synonym detection across 27,000+ open-ended definitions, error-detection for over 9,000 concept maps, and restructured / optimized Excel workflows.
• Cut model training from as long as 48+ hours to under 20 minutes by replacing the lab’s MATLAB neural-network ensemble with a Python ANN that reached R² 0.40 on held-out students.

Python, Rust, TypeScript, C++; giving back to tools I use, and learning how large teams review and ship code

• Merged in XGBoost and Tauri; cherry-picked into DuckDB v2.0; led to a Hugging Face Hub fix (v1.33.0).
• Approved by maintainers in ONNX Runtime and Electron Packager (merge pending); open at PyTorch and Tauri.

Fusion 360, 3D Printing, Arduino, C++, AccelStepper, GoogleTest

• Built a stepper-driven syringe pump: NEMA 17 on an A4988 at 1/16 microstepping, 0 to 7.5 mL/min, LCD and interlocks.
• Rebuilt the Arduino firmware as plain C++ that runs on a PC with the hardware mocked; 40 unit tests, CI on every PR.

Bachelor of Science, Computer Science

• Courses: Algorithms, Data Structures, Operating Systems, Computer Architecture, Software Engineering, Systems for AI, Social Network Analysis (GNNs), Probability & Stats, Machine Learning & NLP (SP27), Networking (SP27), Cybersecurity (SP27)