Case Study: Refuel.ai achieves faster, more accurate data labeling with Databricks

Refuel.ai - Customer Case Study

Refuel.ai wanted to replace slow, error-prone manual data labeling with a faster, more accurate LLM-based workflow. To build its Refuel LLM, Refuel.ai worked with Databricks and Mosaic AI to create a custom model for data cleaning, labeling, and enrichment, with support for large-scale experimentation and training infrastructure.

Using Databricks Mosaic AI Training, Refuel.ai trained nearly 50 models over about three months, then fine-tuned the model for target domains. The result was a 78% increase in label quality, followed by an additional 16% performance gain from fine-tuning, along with faster feedback cycles that could be up to 100x quicker.

Refuel.ai

Nihit Desai

Co-Founder