Reinforcement learning 3

Platform for building, evaluating, and improving AI agents for business automation. Foundry is a platform to build, evaluate, and improve AI agents that can automate key parts of your business—customer support, hiring, sales, and more. Foundry is a platform for building and evaluating browser agents. Users can configure tasks, define evaluation criteria, and collect high-quality data for reinforcement learning (RL) and agent improvement. Our platform provides a deterministic web simulator and an annotation framework so you can collect labels, benchmark agents, and debug performance without worrying about web drift, IP bans, or rate limits.

Predibase is a low-code AI platform for building, fine-tuning, and deploying state-of-the-art AI models, including LLMs. Predibase is a low-code AI platform that enables engineers and data scientists to build, optimize, and deploy state-of-the-art models, ranging from linear regressions to large language models, using minimal code. It specializes in fine-tuning and serving LLMs, allowing users to customize and serve open-source models that can outperform GPT-4 within their own cloud or Predibase's infrastructure. The platform offers features like reinforcement fine-tuning, multi-LoRA inference, and autoscaling infrastructure to ensure unmatched accuracy and speed in training and serving AI models.

AI world models that learn and simulate physical and virtual environments Odyssey is an AI research lab developing general-purpose world models: causal, multimodal systems designed to understand and simulate how the world works. Its models learn about physics, motion, cause and effect, and agent behavior. Odyssey presents these models as a shared foundation for physical and virtual applications, including robotics, autonomous driving, drones, gaming, and AI training. Its offerings include Odyssey-3, Agora-2, PROWL-2, and Starchild-1.