RLDataGym

RLDataGym is a web app that helps teams convert real-world event logs into clean, reproducible offline reinforcement learning datasets. Most RL projects stall before training because data is fragmented across telemetry, product analytics, and backend events, with missing actions, delayed rewards, and inconsistent state definitions. RLDataGym ingests raw logs, guides you through defining state/action/reward schemas, and automatically builds trajectories with versioned dataset artifacts. It flags common offline-RL failure modes (leakage, non-stationarity, action sparsity, reward delay, policy coverage gaps) and produces standardized exports for popular libraries. It also generates a “dataset card” with assumptions, known biases, and evaluation caveats so experiments are auditable. This is an AI app + traditional app combination: AI assists with schema suggestions and anomaly detection, but the core value is deterministic data engineering and governance for RL.