I'm an engineer working on AI safety and security.
I'm interested in what capable AI systems can do in adversarial environments: how agents fail, how to red-team them, and how to build evaluations and safeguards that give defenders the advantage.
- agent-vcs: source control for a world where most clients are agents, from scratch in Rust. Design first, nothing works yet.
Most repos below are archived learning projects and short MVPs. Each README says what it is, what it isn't, and why I built it.
- notebook-parser: photos of handwritten notebooks → tagged Obsidian notes, with Claude's vision API
- llm-server: LLM inference server with structured-output validation, batching and rate limiting
- inbox-search: semantic search over my open browser tabs
- pycones23: my PyConES 2023 talk on ML experimentation with MLflow
Work on AI safety and security is coming. I'll pin it here when it's worth showing.
- Writing: AI by Mai
- Newsletter: AI by Mai on Substack, every two weeks on AI engineering and cybersecurity, starting 2027



