A probabilistic process calculus that models each LLM invocation as a first-class measurable distribution over continuations rather than a deterministic function. Built on six primitives with a probabilistic small-step operational semantics, the calculus enables formal proofs of replay soundness under nondeterminism, compositional probability bounds across concurrent agents, and information-flow guarantees for multi-agent memory access. Accompanied by a zero-dependency Rust reference implementation whose modules mirror the calculus one-to-one.

Published paper presenting a CNN trained for high-accuracy oral cancer detection, with custom hardware for optimal image capture and an open-access application.