Kabam Keeps Game Development Front and Center with Weights & Biases

"Before W&B, whenever there was a behavior discrepancy, it was really hard to figure out what was wrong. Now, whenever we have a problem, we can usually solve it quickly, and most of the time, it’s as simple as looking at our dashboards."
Eric Chou
Senior Software Engineer

Small Screen Entertainment

This year, there will be over 3 billion mobile players worldwide. That’s great news for every video game manufacturer, including Kabam, a leader in immersive mobile games.

But with all those gamers in the world, it’s important to remember: it’s one thing to earn a download of your game, but it’s another thing to keep an audience engaged and keep coming back for more. Bugs or glitches, for example, can often kill the mood for players. Slow loading, crashing, freezing, and graphic or audio issues can escalate from slightly annoying to extremely frustrating.

While creating an error-free game is impossible, the ML team at Kabam strives to come as close as possible.

Shortening Feedback Loops

While we’ve seen ML leveraged in the gaming industry in many ways, such as improving game balance and adding reactivity to non-playing characters, Kabam turns to ML for a different reason: an AI game design assistant. The first objective of the AI system is to improve and build upon the team’s ML research.

Imagine a tool that can help ML practitioners in the same vein as what spell-check did for writers. Sounds pretty great, right? The team at Kabam has made that tool a reality, giving them confidence in their day-to-day work. “The ML pipeline typically has many moving parts—datasets, features, models, parameters—the list goes on. Changes in one component can cause errors in other components,” says Eric Chou, Senior Software Engineer at Kabam.

Take, for example, adding a new piece of code to an existing ML project. The biggest fear for any practitioner is breaking the system. With the solution the team built, they can run the new code against it to see if any bugs will occur. If no issues arise, they can rest assured and continue to the next step.

As news about the effectiveness of the new tool spread across the company, other teams like game design soon leveraged it into their workflow. “Whenever the game team introduces a new feature in a game, they use our system to test it and ensure nothing existing is disrupted,” Eric said. “If something does go wrong, then we can catch it early and decide if we want to push the feature a bit further down the road.”

Tackling Problems Smarter and Faster

While the team’s AI assistant was providing reliability for Kabam, in the background, Weights & Biases also ensured the new tool was behaving reliably.

Using the wandb.log function, the team can log almost everything they want in various ML workflows—numbers, images, video, audio, HTML, 3D, and much more. This allows the team to visualize and debug model pipelines with ease.

“Before W&B, whenever there was a behavior discrepancy, it was really hard to figure out what was wrong. Now, whenever we have a problem, we can usually solve it quickly, and most of the time, it’s as simple as looking at our dashboards,” said Eric.
Past that, for an innovative team like Kabam, efficient experimentation is critical. With the different metrics they can observe in Weights & Biases, the team can easily compare new experiments to previous runs and check for consistency. Having that visibility helps foster a robust culture of experimentation, enabling the team to stay curious and take risks. The creation of the AI assistant is a prime example of a well-executed idea with impactful results.

Changing the Future of Gaming

Known for delivering cutting-edge entertainment in mobile gameplay, Kabam is constantly looking for ways to drive innovation reliably and efficiently. With the implementation of ML in the organization, the team harnessed the power of AI to build a solution that could help them develop dependable ML pipelines.

Supercharging their work with Weights & Biases, Kabam is equipped with a fast feedback loop to quickly identify problem areas and address them. In the fast-moving gaming industry, catching any unexpected behaviors that might cause flaws is critical to ensure a great player experience.
Curious about Kabam’s games? Visit their website to learn more or, better yet, try them out for yourselves