Snorkel AI

Snorkel AI develops expert training data, evaluation systems and runnable environments for frontier AI labs and enterprise teams. Snorkel Flow labels and manages training data programmatically; Snorkel Evaluate helps teams create benchmark datasets, build specialized evaluators and identify error patterns. Its Expert Data-as-a-Service supplies datasets for evaluation and post-training, while Snorkel Data Series packages datasets with difficulty tiers, rubrics and evaluation slices. The company also builds specialized agents for enterprise workflows, using task-specific checks to assess their performance.

Founded in 2019 out of the Stanford AI Lab, Snorkel AI’s co-founders are CEO Alexander Ratner, Christopher Ré, Paroma Varma, Braden Hancock and Henry Ehrenberg. Its research roots include the 2017 Snorkel system for weak supervision: users write labeling functions that express heuristics, and the system statistically denoises their potentially inaccurate, correlated outputs to create training data. That emphasis on defining and measuring data quality extends to collaborative benchmarks such as Senior SWE-bench, which tests coding agents on feature implementation, runtime debugging and adherence to codebase conventions.

In 2025, the company reported production users including BNY, Wayfair, Chubb and the U.S. Air Force, and work with seven of the top ten U.S. banks. It raised a $100 million Series D led by Addition that year at a $1.3 billion valuation, bringing total funding to $237 million.