Game-theoretic AI

Mieza computes provably optimal strategies for domains with hidden information and adversarial dynamics. Where LLMs give advice that is probably good, our solvers compute strategies that are provably optimal -- equilibrium policies from counterfactual regret minimization, translated back into actionable decisions.

The stack: Clojure orchestration, C/OpenCL GPU solvers, custom compression pipelines for game trees at scale. One codebase, production-grade.