Weights & Biases is a developer-tools company whose platform helps machine-learning teams track, visualize, and reproduce their work. It was founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis, who describe starting the company “behind a karate studio in San Francisco with a simple but ambitious goal: to build the best tools for machine learning.” Biewald and Van Pelt had previously co-founded the data-labeling company CrowdFlower (later Figure Eight).
The platform began with experiment tracking, logging the parameters, metrics, and outputs of each training run so a developer can compare runs and see how a model evolved over time, a process that was largely manual and error-prone before such tools existed. It expanded into a broader MLOps suite including hyperparameter sweeps, dataset visualization tables, a model registry, collaborative reports, and Weave for working with LLM applications. Its customers include OpenAI, NVIDIA, and Cohere.
In May 2025 Weights & Biases was acquired by the AI cloud provider CoreWeave, in a deal widely reported at around $1.7 billion, folding the tooling layer into a GPU-cloud business.
Why business readers should care: experiment tracking is the unglamorous record-keeping that makes machine-learning work auditable and repeatable. Tools like Weights & Biases are where teams prove which model version was trained on what data with which settings, which matters for both debugging and compliance.