A generative AI startup working to reinvent how plaintiff firms evaluate legal cases needed accurate and cost-effective training data. Their goal was to teach AI agents to handle legal workflows but they lacked access to high-quality legal datasets and experts who understood the specifics of insurance claims, legal documents, and medical records.
The startup partnered with Databrewery to build a skilled team of professionals through the Brewforce network. These experts had strong backgrounds in legal processes and terminology. The focus of the project was on document-level text extraction and training the AI agent to identify and understand relevant legal information as a human legal expert would.
The startup received high-quality legal datasets with fast delivery and smooth execution. Databrewery managed the project end to end, making it easy for the client to integrate legal knowledge into their AI agents. With this foundation in place, the company is now set up to expand the scope and depth of its legal AI capabilities.
A fast-moving generative AI legal startup set out to build an AI agent that could handle tasks like contract review, risk analysis, compliance checks, and legal document processing work that typically demands hours from legal professionals.
To get there, they needed highly specialized training data. Foundational models lacked the domain-specific reasoning and context required for legal use cases. The company had to develop original datasets built around the nuances of the legal industry, then apply targeted fine-tuning and post-training to teach the model how to perform like a legal expert.
To support the AI agent’s development, Databrewery quickly assembled a team of legal experts through the Brewforce network. These professionals were onboarded to handle complex legal data tasks. At the same time, Databrewery collaborated with the startup to define clear labeling instructions and build a custom ontology that captured every essential data point.
The startup shared a set of prompts linked to multi-page legal documents. The experts were tasked with extracting key insights, writing well-reasoned responses backed by evidence from the documents, and reviewing the model’s output for accuracy, safety, and legal soundness.
A second phase of the project focused on insurance and medical billing documents ensuring the model learned to correctly interpret and extract highly specific industry details.
After testing other providers, the company chose Databrewery for its ability to deliver accurate, high-quality human-labeled legal data fast and for its powerful platform that made it easy to build and manage tailored project ontologies.
With support from Databrewery’s data factory, the startup improved model accuracy and accelerated development, staying ahead in a rapidly evolving legal AI market.