Data Scientist & Creative Systems Engineer — San Francisco Bay Area
Applied data work from 2016 to 2023 across the World Bank, NORC at the University of Chicago, and Rios Partners, where I grew the data strategy work from an informal group into a standing practice and led it — NLP, geospatial, and web-scraping pipelines, and the statistical modeling built on them. I didn't come up through a computer-science program; I came up through public policy and music production, and I still ship the models myself. Since 2024 the proving ground has been audio: neural models running on-device, and a real-time C++ codebase with an allocation-free audio-thread test.
- 📄 CV — zachscheffler.com/cv
- 🌐 Portfolio — zachscheffler.com
- 🔬 Research — research.zachscheffler.com
World Transit Atlas — 201 metro and light-rail systems assembled from OpenStreetMap and open agency data. The interesting part is the measurement: ridership figures use three incompatible counting conventions, and separating them turned a headline result into a smaller, defensible one. The write-up documents the four rounds in which the claim shrank, and the check that found 28 of 30 Chinese systems mislabelled against the national statistics report. → case study
Music Catalog Intelligence — acquisition scoring, regression forecasting and rolling-variance risk over a public Spotify dataset. Every threshold is asserted in tests so the prose cannot drift from the code, and the page states what the forecast band gets wrong rather than hiding it. → case study
Session-State Analyzer — one canonical schema across four DAWs, with partial observability treated as a first-class evidence tag rather than a missing value. → case study
AutoHarm — a generative chord instrument running a four-corpus Markov blend and two ONNX models in the browser, playing live MIDI into a DAW. Ported from a Max for Live device. → try it · case study
Built on the Ableton Extensions SDK (TypeScript):
- Drum Rack Cartographer — analyze, label, preview, and remap drum MIDI into a readable Drum Rack map
- Track Doctor — clean up messy sets with smart, reviewable track renames
- Arrangement Architect — turn Session View sketches into section-by-section song blueprints
- Live Console — a command-palette workflow utility, inspired by REAPER's ReaConsole
Producing since 2013, and an M.M. in Music Production, Technology & Innovation from Berklee in 2025. It is not a side note: audio is the hardest thing I point these methods at, and most of what I know about latency, measurement and shipping under real constraints came from it.
→ Streetcar Scandal on Spotify · session and engineering credits