AI & SOFTWARE
Machine learning systems and software products, from a research thesis to tools people use.

Guardian: danger detection from wearable data
Stop the Spike is building Guardian, software that uses biometric data from existing wearables to detect drug-related harm early enough for someone to step in.
I lead the machine learning work: designing the detection models, organizing and interpreting clinical study datasets, and acting as the technical liaison between internal R&D and outside principal investigators at Stanford Medicine.
- RoleDirector of Applied Machine Learning & Biomedical Analytics, Stop the Spike (2025 to present)
- StatusIn development. Details available on request.
- ToolsPython, time-series machine learning, wearable sensor data
Break Boss: a neural network that finds breakbeats
Producers dig through thousands of records to find a few seconds of usable drums. Break Boss automates the search.
A modified convolutional network, the Perceptron Audio Classifier (PAC), learns the spectral patterns that separate breaks from non-breaks. A second tool, the Break Identification Genius (BIG), uses those results to find the best sections of any sample source. I taught myself Python to build it in nine months.
In 2026 I rebuilt it to run in the browser and tested both versions on records they had never heard.
- RoleResearch, model design and all code. Master’s thesis, University of Miami
- ResultOver 98% in the 2020 thesis test. Rebuilt in 2026: 98.3% on records it had never heard
- ToolsPython, PyTorch, convolutional networks, ONNX Runtime Web

AICA: an AI platform for distribution and category management
AICA set out to give distributors and brands a machine learning view of what to stock and where.
I led development of the platform and its service: the application, the database, search built on machine learning, AI chat, and dashboards that curate content for each user. I also ran marketing, design testing with clients, and data licensing.
- RoleCo-founder, CMO and Head of Digital (2023 to 2024)
- ScopeFull-stack development, database, UI/UX testing with clients, data licensing, small team
- ToolsNode.js, React, Figma, FlutterFlow, Supabase, Docker, GCP
Our Song: a 4D map of people, shows and songs
Streaming services know what you play. They don’t know who you heard it with.
Our Song maps that layer. The first version plots every Pretty Lights set I’ve logged since 2009, the fans in the room and the songs heard most. Drag the timeline to replay it. Two people can pin “our song”, the track that makes them think of each other, and every pin adds a person-to-person-to-song link to the map.
- RoleConcept, design and build
- StatusWorking prototype on real show data, with simulated fans
- ToolsJavaScript, 3D graph rendering, single-file web app


Groundwork: a 36-month forecast for new small businesses
First-time owners sign leases on guesswork, and the usual answer is a static spreadsheet template.
Groundwork asks where you’re opening and what you’ll sell. Local wages, payroll taxes, rent, prices and seasonality come from public data for that area. It returns a monthly P&L, cash flow, the break-even month, lender coverage, a downside case, and a check against what nearby restaurants actually do.
- RoleProduct definition, design and build
- StatusVersion 2 built, in testing
- ToolsTypeScript, React, public wage, rent and price data
AI-built tools
I design the system, then direct AI agents to build and test it. Four that run today:
- Trading brief: pre-open and pre-close market briefs, with a nightly review that paper-trades every pick and keeps score.
- New business filings: one clean feed of new company registrations from four states’ open data.
- Market maker: a paper-trading bot that quotes both sides of prediction markets to earn liquidity rewards.
- Taskboard: a corner window that ranks my tasks by how much unattended work each one unlocks.
- RoleArchitecture and direction, built with AI coding agents
- ToolsPython, SQLite, public data APIs, Claude
Open-source showcase
A library of small, finished apps: planning calculators, API tooling, evaluation harnesses for language models, retrieval from scratch, and more. Each one has a README, tests that pass, and a license, and many run in the browser.
- Result111 public repositories with passing checks, 32 with live demos
- ToolsPython, JavaScript, GitHub Actions, GitHub Pages
