I build the connective tissue between scientific software, reliable delivery, and practical AI systems—from Python packages and release pipelines to local-first agent workflows.
My bias is toward tools that are reproducible, inspectable, and useful outside the demo.
| Focus | What that means in practice |
|---|---|
| Scientific Python | Packaging, PyPI and Conda distribution, Array API work, and accelerated compute workflows |
| Delivery systems | Cross-platform builds, automated testing, release pipelines, and dependable developer tooling |
| Local-first AI | Private-by-default inference, human-supervised agents, and deliberate local/cloud routing |
| Systems engineering | Linux internals, distributed systems, and the layers beneath high-level applications |
|
Local-first mobile bill splitting with OCR and deterministic reconciliation. Python · FastAPI · Ollama
|
Routes work between local models and authority models for cost, privacy, and verification. Shell · llama.cpp · Codex
|
|
Validated agent personas, team templates, and safe deployment helpers with human oversight. Python · Agents · Workflows
|
Advanced systems programming exercises focused on Linux device-driver fundamentals. C · Linux · Systems
|
design constraint → explicit trade-off → repeatable build → observable result
Based in the United States · Building in public since 2018
