Case Study: Eve builds reliable legal AI agents with Braintrust

eve builds reliable legal agents with Braintrust and 400+ plaintiff law tasks

eve, a legal AI company for plaintiff lawyers, needed to scale the evaluation of its AI agents beyond manual reviews to ensure reliability across complex tasks like document analysis and legal research. They partnered with Braintrust to build a systematic way to measure performance and catch fine-grained errors before they reached customers.

Using Braintrust, eve built Plaintiff Bench, a gold-standard benchmark dataset with over 400 expert-graded legal tasks tagged by failure mode. This allowed them to test model changes rigorously, automate the evaluation process with coding agents, and turn production failures into reproducible test cases. As a result, eve can now run multiple agent improvements in parallel with confidence, leading to better performance on key metrics like citation coverage and instruction following for their customers.

eve