Training AI Like an Athlete: The Story of Our Legal Tool

According to fitness experts, there’s no one moment when you “get in shape.” There’s no threshold that you can cross where you can declare yourself “in shape,” but there are a series of milestones such as being able to run five miles or being able to bench press 300 pounds. Those goals are different for everyone and depend on how you define your own fitness. The same is true in the journey to build our legal AI. Some products come from a single eureka moment, but others are the culmination of the achievement of multiple milestones until the product is “in shape.” Building our legal AI was a journey that started with a technology concept, found a product market fit, and iterated until something extraordinary was created. Our legal AI is privacy preserving artificial intelligence (AI) for legal professionals that delivers a time saving, efficiency building, and trust inspiring AI capabilities for multiple legal organizations. Getting there was an exercise in listening. It was an exercise in recognizing that we had a technology that we needed to build into a product. Much like the pursuit of fitness, it is a journey without an end. We’ve built something we are proud of and with which we’ve seen success in the market. But as the COO, I’m prouder still of the process and how our people and our culture came together to build something extraordinary for an important market. Origins The roots of our legal AI goes back years. My cofounder and I worked in federal government positions prior to creating Frontier Foundry. Both of us worked in positions where we had access to data that was restricted in one form or another and we both came to the same observation. Those with access to the most sensitive datasets are often those that do not have access to the best tools. Technology innovation thus far had excluded those that were not willing to transmit their data to cloud parts unknown. The market had accepted that these tools were not built for privacy and that using them meant that you by default had to accept a lack of privacy and the constant threat of data leakage as a feature, not a bug. Having been in the chair with data that was restricted and having to use tools from decades prior, this did not seem like something we should accept. This was the challenge that led to creation. Over the next year, Frontier Foundry found the best engineers and gave them a hard problem: build industry leading AI, but you don’t get to use a massive cloud. Every piece of technology needed to run on a mid-range laptop. This forced us to be more efficient with how we built and eliminated the potential for lazy engineering that comes from access to nearly endless computing power. When all you have available is a laptop’s worth of compute, you must be smart . The intent was not to give customers AI that could ONLY work on a laptop, but to demonstrate that AI capabilities, such as large language models (LLM) could be built, trained, and operated in environments secured from the outside internet. The real application would be to deploy the models inside a customer’s secure environment, but the laptop functionality ensured strong engineering and demonstrated the capability. With the technology in hand, we began looking for markets. Fortunately, there are several markets with demand for AI that preserves privacy, so opportunities came up from multiple angles. However, one market stood out because of the amount of incoming interest that we fielded, even without dedicated marketing efforts: legal. After fielding half a dozen calls from legal organizations and signing a few on as customers, we looked at each other and had the eureka moment. It seemed so obvious once the market came to us. We were able to demonstrate some early success in criminal law and contract law with customers proving the concept. To review the fitness journey of our legal AI: A proven technology concept. Incoming market interest. Proof of concept in the legal field. While our legal AI was clearly getting in shape, we needed to hit more milestones. With product market fit clearly in sight and demonstrated technology, we needed to move toward productization. This is where the true heavy lifting began. The ensuing months were filled with reps upon reps of research that challenged even my graduate school years. We read, we interviewed, we bought coffees…all in pursuit of understanding exactly what problem we were solving. One of the saddest places to be in technology building is to be considered a solution in search of a problem. We knew that was not our case, but we still had to work to build for an industry that has very specific needs and restrictions. Fortunately, our work with restricted data in the past prepared us well for the challenge. The restrictions were different, the language was different, but the goal was the same. Productization Researching a market can feel like an academic exercise, but there was a moment in our work to build AI specifically for the legal market that broke us out of the academic feelings. One of the reasons that the data used by legal professionals is so restricted is because the stakes are so high. Peoples’ lives or livelihoods are on the line in legal cases build from a foundation of the Constitution. This is said not be sound high minded but to state fact. AI for the legal profession is fundamentally different than other AI. AI built for general purposes such as OpenAI, Microsoft CoPilot, and Anthropic’s Claude might be serviceable for some applications, but they also lack the specific privacy, security, and model specialization that legal entities require. This realization gave us the drive to push into a serious market niche with implications for how legal professionals do their work. The Result Our legal AI is specialist AI for the legal industry. It was built with security and privacy first. As with all Frontier Foundry AI, our legal AI was built to increase productivity of individuals. Our legal AI tackles the most time consuming and detail-oriented tasks for the most document heavy cases across the legal industry. We built in features like: Audio and Video Transcription Automated Ingestion and Training for Case Documents Secure LLM Interface that Does Not Require Cloud Connectivity Optical Character Recognition for Handwritten Documents Customized Drafting Connectivity to External Resources Here’s where I would say “but wait, there’s more” if our marketing analyst hadn’t rolled his eyes so dramatically when proofreading (true story-note from Thomas, the marketing analyst) . We also built in auditability and transparency features that legal professionals need when presenting findings or strategy to clients and courts. Our system has the ability to log queries and outputs per user and give supervisors an enterprise view of how the tool is being used. This feature is specifically designed to give users the cover they need to use the system confidently. Knowing where information came from (via our citations feature) and how someone got to it (via our query logging feature) creates AI that is trust inspiring and gives firms the capabilities they need to keep up at a time when cases have more and more data and the expectations from clients are ever higher. This system delivers more accurate results and allows firms to reduce pass-through costs while bringing on more cases to drive higher profit margins. Building out legal AI was like building muscle or endurance in fitness. It didn’t happen all at once. It started with a foundation and grew as the vision became clearer. Our team hit milestones one after the next until we arrived at a true product that is already impacting legal firms and how they use AI. The best tools should not be off limits to those doing the most impactful work. Frontier Foundry was built on that premise and our legal AI is the latest reflection of it. As we continue to innovate, we are finding new ways to preserve privacy but equip attorneys with the best tools. Just as fitness does not happen all at once, it also must be constantly built and maintained. That’s our legal AI. Connect with us: Substack , LinkedIn , Bluesky , X , Website To learn more about the services we offer, please visit our product page. Nick Reese is the cofounder and COO of Frontier Foundry and an adjunct professor of emerging technology at NYU. He is a veteran and a former US government policymaker on cyber and technology issues. Visit his LinkedIn here . This post was edited by Thomas Morin, Marketing Analyst at Frontier Foundry. View his Substack here and his LinkedIn here . Subscribe now Leave a comment