refinery

Why we included this project

For data scientists building NLP models, the labeled data is often the real bottleneck, not the algorithm. Refinery gives you a web-based workspace to annotate text, review and correct label quality, and keep a record of how each dataset was assembled. It builds human-in-the-loop into the flow, with active-learning suggestions pointing to the examples that most need labeling and automated checks that flag disagreements between annotators. If you manage multiple annotators or iterate on dataset versions, this turns label management from exported CSVs into a repeatable, auditable process.