I'm building the infrastructure that lets AI and scientists do research together without the work quietly corrupting itself. I care about provenance, an honest accounting of what's actually been verified, and the trustworthy substrate underneath the agents.

Software has tests; biology mostly doesn't. Point an optimizer at a goal with no honest verifier, and it gets good at plausible, fluent, quietly wrong answers faster than anyone can check them. Closing that gap is the work.