Use Cases

Cobalt is designed to help you understand, compare, and debug machine learning models and data. Below are some of the most common use cases.

E-Commerce Embedding Model Comparison

Choosing the right embedding model for a vector database is critical for search quality. Cobalt clusters users’ product search queries into interpretable categories and generates a model comparison table, making it easy to evaluate which embedding model performs best for your use case.

See the E-Commerce Embedding Model Comparison notebook for a hands-on walkthrough.

Fine-Tuning Tradeoff Analysis

When fine-tuning an embedding model, it’s important to understand what you gain and what you lose. Cobalt reveals performance tradeoffs between a base model and its fine-tuned version through the model comparison table, helping you make informed decisions about whether fine-tuning is worth it for your data.

Text Classification Debugging

Cobalt can be used to explore and debug transformer-based text classification models. By examining model behavior across automatically discovered subgroups, you can identify systematic errors and understand where your model struggles.

Data Drift Diagnosis

Even without a model or performance metrics, Cobalt can help diagnose and understand data drift by comparing the structure of two datasets and surfacing meaningful differences between them.

Image Clustering and Exploration

Using embeddings from models like CLIP, Cobalt can help you make sense of image datasets by discovering natural clusters and exploring how target labels relate to the underlying structure of your data.

---

For runnable examples of each use case, see the Example Notebooks or browse the notebooks on GitHub.