Snorkel AI

Snorkel AI is a data-centric artificial intelligence company founded in 2019 in Redwood City, California, United States, spun out of the Stanford AI Lab. It was created to solve the bottleneck of manually labeling and curating the data needed to train and evaluate specialized AI models.

The platform lets enterprises programmatically build and manage training data, then use it to develop and evaluate specialized AI agents against task-specific acceptance criteria. Snorkel AI also publishes benchmarks such as Terminal-Bench and Senior SWE-bench, and works with frontier AI labs including Google, AWS, Microsoft and Mistral AI on data development for their models.

1. Programmatic labeling: converts subject-matter expert knowledge into labeling functions instead of manual annotation.

2. Specialized agent development: builds custom AI agents evaluated against defined task criteria.

3. Expert-curated datasets: structures data with acceptance criteria and designated verifiers.

4. Benchmarking suite: offers Terminal-Bench, Senior SWE-bench and OSWorld for evaluating agent performance.

5. Frontier lab partnerships: collaborates with major AI labs on data development for large models.

6. Domain benchmarks: extends evaluation to coding, legal and insurance use cases.