I lead AI & data engineering in drug development by day. This is my personal account — personal projects, open-source history, and experiments.
Currently exploring agentic coding infrastructure:
- project-sandbox — running coding agents in isolated containers and microVMs, with supervised or headless sessions, LLM-gateway credential isolation, and git-worktree / Jujutsu workspace integration
- agentgateway-locally — keeping credentials out of the agent container, and using locally hosted models
- jj-parallel-agents-skill — parallel agents on jj repos, plus an experiment in testing agent skills with evals
- repljail — an MCP tool for network-isolated R sessions
Previously: genomics methods and benchmarking at Illumina — hap.py and Paragraph.
More on my blog