Product leader + AI builder

I build AI systems from research and prototype through shipped product.

Founder and hands-on builder working across local inference, agent infrastructure, developer tools, and human-AI interfaces.

View selected work Say hello

Selected work

Products where I owned the problem deeply enough to design, build, and measure the system.

threadfork local-first AI workspace showing a launch decision transcript and cited assistant answer

threadfork · Founder / Product / Engineering

Private AI that runs where your conversations happen.

I designed and built a local-first desktop workspace that turns conversations into private, searchable context using on-device speech and language intelligence.

RustTauriWhisperllama.cppMLX

On-device transcription, diarization, retrieval, entity extraction, and local LLM inference.
Human-AI workflows for branching, synthesis, comparison, and durable knowledge.
Shipped desktop product with early paying customers and design-partner feedback loops.

Visit threadfork ↗

DataFog Claude Code PreToolUse hook denying a tool call after detecting credit card, email, and phone data

DataFog · Founder / Open-source maintainer

A fast, local PII firewall for AI agents.

I built an open-source developer platform that detects sensitive data and enforces privacy policy directly inside agent and LLM workflows.

PythonGLiNERNERLiteLLMClaude Code

Matched PII benchmark120–171× measured

1.2 KB denseDataFog 0.23 ms · Presidio 27.7 ms

120×

10 KB mixedDataFog 1.29 ms · Presidio 186 ms

144×

100 KB sparseDataFog 12.2 ms · Presidio 2,090 ms

171×

EMAIL · PHONE · CREDIT_CARD · SSN

Benchmark methodology

Medians from the reproducible benchmark suite, measured on identical payloads and entity types.

Visit DataFog ↗

Lab

Experiments in AI systems, inference, agents, and machine learning.

Reinforcement learning

Tesserack

Compiling strategy guides into dense, interpretable reward tests that train game-playing agents.

Reinforcement learningClaude VisionWebGPU

View on GitHub ↗

Inference

Optimizing local LLM inference on Apple Silicon

Experiments across llama.cpp, MLX, decoding, memory, and on-device model performance.

MLXllama.cppApple Silicon

Modeling

Time-series transformers

Built a benchmark-beating regime-aware transformer that fuses six weather-model feeds with live station observations to forecast Central Park’s daily high.

TransformersEvaluationTime series

Accessibility tooling

TerminalDeck

A hands-free coding interface for controlling Claude Code and Codex with Stream Deck and voice dictation.

TauriRustStream Deck

View on GitHub ↗

About

I’m a founder, product leader, and hands-on software builder. Before starting DataFog and threadfork, I led cross-functional teams building digital health platforms, machine-learning recommendations, and internal data products.

I tend to work on problems where making the right product decision requires understanding the underlying system well enough to build and test it.