Provides human‑curated training datasets, agent environments, and validation tooling to improve reasoning and real‑world task performance of large foundation models