Active Learning Task Prioritization And Uncertainty Sampling

via “task sampling and active learning queue management”

Open-source multi-modal data labeling platform.

Unique: Decouples sampling strategy from task storage via a pluggable algorithm interface that accepts external ML predictions, allowing teams to swap sampling strategies (random, sequential, uncertainty, consensus) without modifying core task models or database schemas.

vs others: More flexible than Prodigy's built-in active learning because strategies are pluggable and can combine multiple signals (model confidence + annotator disagreement); more lightweight than Snorkel because it doesn't require training weak labelers, only ingesting predictions.