Upload your dataset to get a health check of its quality, quantity, and diversity. Zoom in and out of your dataset. Uncover distribution biases before you train. Find and fix labeling errors quickly.
Upload model inferences against your labeled datasets and deep dive into its performance. Find where your model is performing well and badly so you can take the best actions to improve it.
With knowledge of your dataset diversity and model performance, Aquarium automatically samples the best data to sample to label and retrain on. Your model performance just gets better.
How it Works
Most model improvement comes from curating high quality datasets. Send us your datasets and your model inferences through our API and we'll find ways to improve your datasets and model performance. Our embedding search technology then helps you sample the best data to label to improve your model, giving you the most model performance improvement for the least labeling cost.
See Aquarium in action
Find failures in your model performance and fix them by intelligently sampling data.