Overview
Aquarium Learning: Embedding-based data retrieval
Aquarium Learning is a data-centric developer platform created to help machine learning teams improve model performance by systematically identifying and correcting data issues. Designed primarily for computer vision and natural language processing applications, the platform automatically surfaces fine-grained failure modes, edge cases, and labeling errors within large training datasets. By utilizing advanced AI retrieval technology and embedding-based search, developers can pinpoint specific data distribution gaps, prioritize the most informative samples for re-labeling, and continuously evaluate performance regressions. The system integrates directly into existing ML pipelines to streamline data curation, active learning, and dataset versioning, eliminating tedious manual inspection. By focusing on data quality rather than tweaking model architectures, engineering teams can achieve faster iteration cycles and build more reliable artificial intelligence systems. Although the platform technology was recently acquired by Notion to enhance enterprise generative AI features, its underlying framework set new standards for automated model diagnostics, dataset debugging, and data curation workflows across production engineering teams.