Disabled veterans seeking to file claims with the Veterans Administration are faced with multiple hurdles and reams of paperwork. Many vets resort to paying third-party companies thousands of dollars to help them with the process. 

What if there were a way to streamline the claims process—to condense burdensome information gathering and data inputting into a linear, simplified set of tasks guided by a chatbot? How long would it take to roll out a tool that could accomplish that? 

The answer: about 48 hours—at least for an interdisciplinary team of students from Stanford University’s schools of Law, Business, and Computer Science collaborating feverishly during Codex’s Large Language Model (LLM) Hackathon held recently on campus.  

The team’s LLM tool, which they named Vet’s Claim, took the award for “Best Overall” at the third CodeX Hackathon, held in conjunction with the 11th annual FutureLaw conference in mid-April. CodeX, co-directed by SLS and the Stanford Computer Science department, focuses on the research and development of computational law, the branch of legal informatics concerned with the mechanization of legal reasoning. 

Vet’s Claim co-creator Camila Chabayta, JD ’25, was one of two SLS students who were part of award-winning teams at the hackathon. Kevin Yan, JD ’26, won the “Best First Build” award along with his business, engineering, and linguistics teammates. His team’s generative AI tool, dubbed DueDiligent AI, automates M&A due diligence related to key contracts and business value assumptions.

CodeX launched the first LLM hackathon last year, just after the 2023 FutureLaw conference, marking the first time that a hackathon was held, anywhere in the world, at the intersection of LLMs and the law. This year, approximately 400 people from four continents participated, including law students and other students from across the Stanford campus—almost double the number from 2023. Over two days, including a day of hands-on training (“the bootcamp”), interdisciplinary teams competed in technical and non-technical tracks to come up with new ideas for streamlining, automating, and augmenting legal practice through the use of LLMs like ChatGPT. CodeX Associate Director and Fellow Megan Ma and CodeX Fellow and SAP Chief Operating Officer Jay Mandal co-led the bootcamp and hackathon. 

“The passionate individuals who participated in the hackathon will likely be future leaders and contributors to the explosion of generative AI applications in the legal space, so it was extremely exciting to witness more than 70 teams creating LLM solutions that completely reimagine the practice of law,” Mandal says. “From the start, the halls and rooms were brimming with intensity and hacker camaraderie.”

Getting Veteran Disability Claims Right the First Time

“Approximately 75 percent of veterans’ applications for disability benefits are rejected the first time they are submitted,” says Houston, Texas native Chabayta, who decided to enter the hackathon after attending a number of CodeX events that sparked her interest in the promise of AI to narrow the access to justice gap and democratize legal practice. Claims are rejected for a variety of reasons, she says, often because of unclear instructions or technicalities around requested information. “We used the latest regulations, 20 years of VA case findings–which are all publicly available online–and AI tools to design, write, and provide a tailored claim that helps veterans get their application for disability benefits right the first time.” 

It helped to have a military veteran on the team: Graduate School of Business student and Army veteran Matthew Champion. 

“I have personally navigated the VA benefits program and I learned first hand how challenging, emotionally demanding, and time consuming the process is,” says Champion, an Army Aviation Officer who received a  Bronze Star and Air Medal with a Valor for his service in Afghanistan. “It took me 26 in-person meetings, and over three years for my application to be accepted. We wanted to help the 18 million veterans that are eligible for first-time benefits have an outcome and experience that is easier, faster and better than mine.”

Chabayta, who will spend this coming summer working in the Palo Alto office of law firm Wilson Sonsini, says she plans to continue honing and expanding her passion for AI legal tools once she graduates. Meanwhile, she’s gaining valuable experience networking with venture capitalists. A number of VCs attended the hackathon and expressed interest in potentially funding the real-world roll out of Vet’s Claim. 

“It is all still sinking in and is so much more than I even expected going into the hackathon,” Chabayta says. 

Automating M&A Due Diligence

Yan won “Best First Build” award, along with his teammates, for DueDiligent AI, their generative AI tool that automates M&A due diligence related to key contracts and value assumptions. In seconds, it can identify excessively risky contract provisions, such as short termination clauses and provisions that could permit large fluctuations in cost. It also flags current events that could undermine a business’s financial outlook, such as a manufacturing plant’s closure recently reported in the news.

Yan, who is on leave from medical school studies at the University of Pennsylvania while he pursues his law degree at SLS, signed up for the hackathon to further his interest in how AI is impacting law and technology—despite the fact that he also had a big research paper to write over the same weekend. “I slept some, but not a lot,” says Yan, a native of Cedar Knolls, NJ,  who plans to practice psychiatry while also working in government or health care administration to address health policy issues, including in the areas of mental health and substance use. 

Listen how SLS students on a CodeX Hackathon team have created an AI tool designed to help veterans apply for disability benefits on NPR’s Morning Edition.

The hackathon was “intense, exhilarating and deeply collaborative,” he says. “I got interested in AI and tech after seeing how these innovations could radically transform the healthcare field as a medical student, cutting administrative burden and improving the patient experience.”

While a number of AI tools already exist in the realm of M&A practice, Yan says, DueDiligent AI was designed to address a specific aspect of due diligence not already covered by existing tools.

CodeX’s Ma says the gap-filling nature of DueDiligent AI was part of what caught the judges’ attention. The tool offers a different approach from other legal AI tools in the M&A space, she says.  “DueDiligent AI project is highly specific to the process of identifying and verifying the business valuation of a seller. It allows a buyer to have a more confident scoring of the seller’s current contracts and projected revenue. This is a different approach from other legal AI tools that focus on contract review, rather than the role and risk of the contracts.”

One of the most valuable aspects of the hackathon, Yan says, was the experience of working with an interdisciplinary team. “I learned so much more than I was even expecting to learn from the hackathon and much of that came from my teammates. It was a really powerful experience to work so closely and intensely with this small group over two days. It taught me a lot about cooperation and collaboration.”