Catalog

Search for labeled and unlabeled data using filters for metadata, model inferences, and other attributes like embedding similarity. No need for an engineer to write one-off query scripts just to find data. Send data directly to a labeling project in just a few clicks.

Not all data impacts model performance equally. Through our active learning workflows and uncertainty sampling, you can filter for data with low-confidence predictions to curate and label the right data–not just more data.

Assign custom metadata in bulk to assets that meet Catalog filter criteria, including embedding similarity, without manual labeling. Leverage this metadata in downstream workflows like helping curate data for labeling projects or structuring custom review workflows. Automatically apply this metadata to new data that meets the filter criteria.

View a detailed class distribution of ground truth labels or model inferences to get a better understanding of your data. See how performance metrics like F1 score vary across your data so you can make the most informed decisions when curating data to label.

Not all data impacts model performance equally. Leverage your data distribution, model predictions, model confidence scores, and similarity search to curate high-impact unlabeled data that will boost your model performance.

Don’t let searching for data and edge cases slow your team down or hold up conversations with stakeholders or customers. Instead of relying on one-off query scripts, search and discover data faster inside Catalog.