Cuneyt Ozturk

I work on PRML, an open specification for tamper-evident pre-registration of AI evaluation criteria: the metric, comparison rule, threshold, dataset hash and seed are serialised and hashed before an evaluation runs, so a third party can later check that those exact criteria existed no later than that time and have not changed since. I wrote the specification and all four reference implementations (Python, JavaScript, Go, Rust), the public registry, and integrations with MLflow, Giskard and UK AISI's Inspect. Falsify OU (Tallinn, Estonia) is the commercial vehicle. Self-funded since April 2026.

Personal Website

https://falsify.dev/

Projects

PRML — tamper-evident pre-registration of AI evaluation criteria

Team Member

No grants recorded.