I enjoy helping customers get to the right AI solution: understanding the problem they're actually trying to solve, working through the options with them, and seeing it land. Outside of work, I like building things to learn and experiment. Lately that's been AI security and agent observability, and right now I'm curious about LLM inference.
Comfortable across the full arc — research and engineering through to product, pre-sales, and commercial.
I care a lot about building AI responsibly. Reading Weapons of Math Destruction in an undergrad philosophy class is a big part of why I went into AI in the first place, and it's shaped my path since: mission-led startups, B Corps, and the AI4Good Lab at Mila.
Adversarial fuzzer for MCP servers — tool poisoning, shadowing, and rug-pull attacks. ~91% MCPTox coverage. Deployable as a CI/CD security gate.
Agent observability — intercepts live runs mid-flight, forks from any step, flags behavioural drift. Framework-agnostic.
Turns a memory or feeling into a professional fragrance composition. RAG · custom NER · scent classifiers.


