ML + Data Science

ML + Data Science

Work in healthcare, signal processing, machine listening, and ecological audio.

Background

Applied machine learning and signal processing since 2012, across health, music, and ecological audio. Published at ISMIR, ACM SIGIR, and in diabetes research; 33 granted U.S. patents.

Available for consulting on machine learning, signal processing, and modeling problems, especially in health and sound: prototyping, validation, and model review.

Patents and Publications

Selected Work

  • Tidepool (first on staff, now consulting): data science and algorithms for FDA-cleared insulin dosing. Designed the insulin and carbohydrate metabolism simulator Tidepool uses (currently designing a new insulin model for it) and the sensor-noise safety analysis for the FDA 510(k) clearance of Tidepool Loop, an automated insulin dosing app; also verification and validation, settings optimization, and exercise modeling with researchers at Stanford and UCSB.
  • TYGER, a gaming music service that adapts licensed songs to gameplay in real time: composition algorithms that build from existing recordings, so musicians' playing stays in the music.
  • Gracenote: helped build machine listening technology used by Apple, Spotify, Google, and Amazon, including training the Mood 2.0 models; later consulted on machine learning modeling.
  • U.S. National Park Service: won a competitive bid to build bird detection for field recordings in Alaska.
  • Change.org: recommendation systems connecting people to causes and campaigns.
Healthcare ML Code texture from InsuLearner development

InsuLearner

Open-source algorithms and tools for computing insulin pump settings, with practical integrations for Tidepool and Nightscout data.

Open Project
Ecological Audio Cameron Summers beside an acoustic recording station in Denali National Park

U.S. National Park Service Acoustic Discovery

Machine learning bird-detection tooling for ecological recordings developed with the U.S. National Park Service.

Open Project
Machine Listening Gracenote Mood 2.0 parallel classification architecture diagram

Gracenote Mood 2.0

Archive of my work on deep-learning-based mood classification and music intelligence systems at scale.

Open Project
Learning Tool Screenshot of the interactive precision-recall tool showing predictions at a threshold and the live precision-recall curve

Precision-Recall Visualization Tool

Interactive explainer for thresholding, tradeoffs, and classification thinking.

Open Project

Working Together

For consulting on machine learning, signal processing, or modeling, especially in health and sound, send a note with a little context about the problem.

Contact Patents and Publications