LexisNexis Builds Customizable AI Assistant, Protégé, Using SLMs and Model Distillation
Legal research company LexisNexis is taking a strategic approach to integrating artificial intelligence, opting for a customizable AI assistant, Protégé, that leverages the power of both large and small language models (SLMs). The goal is to provide legal professionals with a tool tailored to their specific workflows, unlike a generalized AI assistant.
Protégé is designed to assist lawyers, associates, and paralegals in writing and proofreading legal documents, ensuring the accuracy of cited materials. LexisNexis aimed to create an AI that learns the nuances of a firm’s processes and provides a more adaptable solution. Jeff Reihl, CTO of LexisNexis Legal and Professional, explained to VentureBeat that the company is using a multi-model strategy to achieve this.
“We use the best model for the specific use case as part of our multi-model approach. We use the model that provides the best result with the fastest response time,” Reihl stated. “For some use cases, that will be a small language model like Mistral or we perform distillation to improve performance and reduce cost.”
Leveraging Small Language Models and Model Distillation
While large language models (LLMs) still offer substantial value in AI applications, some organizations are turning to SLMs or employing distillation techniques, which involve training smaller models using LLMs. This allows for the creation of more efficient and cost-effective solutions. Distillation is proving to be a popular choice for many companies. Small models work effectively for applications like chatbots and streamlined code completion, which aligns with LexisNexis’s goals for Protégé.
LexisNexis has a history of incorporating AI into its products, even preceding the launch of its legal research hub, LexisNexis + AI, in July 2024. Reihl noted, “We have used a lot of AI in the past, which was more around natural language processing, some deep learning and machine learning. That really changed in November 2022 when ChatGPT was launched, because prior to that, a lot of the AI capabilities were kind of behind the scenes. But once ChatGPT came out, the generative capabilities, the conversational capabilities of it was very, very intriguing to us.”
Multimodal Approach with Fine-Tuned Models and Model Routing
LexisNexis utilizes a variety of models from leading providers, including Claude models from Anthropic, GPT models from OpenAI, and a model from Mistral within the LexisNexis + AI platform. This multimodal strategy enables the platform to efficiently manage various tasks.
To facilitate this, LexisNexis designed its platform to switch between different models. Reihl elaborated, “We would break down whatever task was being performed into individual components, and then we would identify the best large language model to support that component. One example of that is we will use Mistral to assess the query that the user entered in.”
For Protégé, the company prioritized speed and specialized models tailored for legal applications. This led them to use “fine-tuned” model versions, which are essentially smaller, more optimized models or the result of model distillation. Reihl noted, “You don’t need GPT-4o to do the assessment of a query, so we use it for more sophisticated work, and we switch models out.”
When a user poses a question to Protégé regarding a specific case, the system first engages a fine-tuned Mistral model “for assessing the query, then determining what the purpose and intent of that query is” before transferring to the optimal model for task completion.
Reihl mentioned that the next model might be an LLM to generate new queries for the search engine or another model that summarizes findings. Although LexisNexis primarily relies on a fine-tuned Mistral model at present, they also incorporated a fine-tuned Claude model initially. LexisNexis is also considering OpenAI’s reasoning models and potentially Gemini models from Google.
LexisNexis bolsters its AI platforms with its knowledge graph for retrieval-augmented generation (RAG) functionalities, especially for Protégé which may help with agentic processes later.
Enhancing Legal Workflows with AI
Even before generative AI’s rise, LexisNexis was investigating the potential of chatbots in the legal sector. The company’s Protégé sits within the LexisNexis + AI platform, offering a set of AI services designed to streamline various legal tasks.
Protégé assists law firms with tasks that paralegals or associates typically manage. It supports the creation of legal briefs grounded in a firm’s data, offers suggestions for legal workflows, refines search prompts, drafts questions for depositions, links case citations for accuracy, generates timelines, and summarizes complex legal documentation.
“We see Protégé as the initial step in personalization and agentic capabilities,” said Reihl. “Think about the different types of lawyers: M&A, litigators, real estate. It’s going to continue to get more and more personalized based on the specific task you do. Our vision is that every legal professional will have a personal assistant to help them do their job based on what they do, not what other lawyers do.”
Protégé now competes with other AI-powered legal research and technology platforms, such as Thomson Reuters’ CoCounsel, which utilizes a customized version of OpenAI’s o1-mini-model and Harvey, backed by investors including LexisNexis.
