Case Study: Devsisters accelerates game development and debugging with Weights & Biases

Devsisters Levels Up Game Development with W&B

Devsisters, the global game company behind Cookie Run, needed a faster, more reliable way to balance gameplay, catch hard-to-find bugs, and keep development cycles moving in a highly competitive mobile gaming market. The team was using reinforcement learning to test and optimize its upcoming title, Cookie Run: Witch’s Castle, and turned to Weights & Biases (W&B) to better monitor model behavior and development progress.

With Weights & Biases, Devsisters gained powerful visualization and tracking tools to observe changing parameter distributions, debug model performance, and monitor hardware utilization in real time. The result was a much faster iteration loop: some development cycles dropped from 2–3 days to just 3–4 hours, and the RL system could train 50 stages in a single day, helping Devsisters launch games faster and with greater confidence.

Devsisters

Yunyol Shin

Applied Data Scientist