7 projects
manifold-cov
Functional coverage and constrained-random verification for AI agents.
crucible-ttc
A verifier-guided reasoning engine: search (best-of-N, beam, MCTS) over reasoning traces to turn test-time compute into measured accuracy.
bonafide
A calibrated, provenance-first framework for detecting AI-generated media.
congruent-eq
Prove behavioral equivalence between an original function and an AI-rewritten one within bounds, or return the concrete input that breaks it.
kinescope
rr for nondeterministic AI agents — deterministic record-replay and counterfactual branching.
toroid
An agent that synthesizes formal properties for RTL, runs a model checker, and iterates on counterexamples.
gatecheck
Provider-agnostic LLM evaluation harness: task suites, deterministic + LLM-as-judge graders, regression gating, and cost/latency tracking.
