Built on game engine technology and engineered for shipping velocity, WorldQL is the collaborative engine for reinforcement learning environments
Build and ship RL environments faster
Multiple people and agents can edit the same environment at once, like Google Docs
Built on web technologies, Dreamlab is easy to deploy and works for all types of applications
Build, test, and deploy environments in hours instead of weeks
Analyze session replays to discover issues early in training runs
Robotic Arm Manipulation Task
“It looks like the model is making high speed random movements in an attempt to achieve the goal by chance. To address this, I'll add a cost per movement to the reward function…”
Simulated CRM / Salesforce Clone
“Viewing this session replay, it appears the model is clicking the submit button rapidly to count the same customer update action multiple times. Fixing this race condition in the UI…”
Track every change, branch experiments, and collaborate easily. Version control built for iterating on complex environments.
See how your environment changed over time
Try new ideas without breaking your main environment
Restore any previous version instantly
“One of the best version control UIs I've seen in a browser”
Our engine is built on web technologies, so RL environments can include anything you can run in a browser. For more complex scenarios, WorldQL includes the Rapier physics engine, a key-value database, and multiplayer support.
We specialize in environments for training domain specific computer use models. We also can provide RL environments for generic coding models focused on game development, one of the hardest and most accuracy sensitive domains of software.
We help labs build custom environments and datasets at scale. We work with a network of over 10,000 registered developers to deliver great results fast.
Send us an email and we'll get back to you within a few hours to discuss your project. Weekends and holidays included :)
You can also book a call.
