For teams building AI in healthcare, finance, or other regulated industries.
Independent review of where a model's output is trusted without being checked, and what a real verification layer around it would look like. Not a checklist audit: a working assessment of where trust is currently unearned.
Rust and z3-based proofs: panic-freedom, safety envelopes, and other properties that should be proven rather than tested for. Built on the same tooling used in cargo-vouch and verified-safety-shield.
EU AI Act and FCA-adjacent work: provenance-gated citation systems, refusal-first design, and documentation that holds up under audit rather than just reading well.
The EU AI Act's technical requirements — logging, human oversight, traceability — describe things a system mus…
The FCA's approach to AI is principles-based: firms need to be able to explain how a model works and demonstra…
Clinical AI has an asymmetric safety profile: a false negative can delay treatment, a false positive can trigg…
Rust's borrow checker prevents memory errors, but it doesn't prove functional correctness, panic-freedom, or c…
Most LLM deployments are trust-based: you trust the prompt, trust the retrieval, trust the output parser. When…
Rate is scoped per engagement. Building Velkron full-time, so this is a small number of engagements alongside that, not a full-time practice.