LexMed uses large language models to help Social Security disability attorneys collect evidence, match medical findings to regulations, and find errors in cases.
2
Nick Coleman built the company from his experience handling about 300 disability cases a year and searching thousands of pages of medical records for evidence.
3
Hearing analysis can flag errors in vocational testimony and help attorneys appeal cases without listening to every hearing recording or manually checking every job claim.
Summary
Nick Coleman explains how his work as a Social Security disability attorney led him to build LexMed. Attorneys must assemble evidence from large medical records, match clinical findings to detailed regulatory requirements, and prepare cases for possible appeals. Nick found that manual search tools could not handle the many different medical terms used for the same finding. Large language models made it possible to apply his decision-tree logic to this work. LexMed is also analyzing hearing transcripts, identifying speakers, correcting recognition errors, and checking vocational testimony against job databases and regulatory rules. The goal is to reduce the time attorneys spend on repetitive review so that contingency-fee practices do not have to choose which clients receive deeper case preparation. Nick is direct about the limits of the system: it supports legal judgment and record development rather than replacing the attorney.
Social Security disability cases depend on evidence collected across a difficult government process
Nick describes Social Security disability as a safety net for people who can no longer work, but qualifying requires proof of a disabling condition expected to last 12 consecutive months. The condition can be mental or physical. Attorneys collect medical evidence, information from employers, and explanations of why a claimant cannot perform past work or an easier job. Applicants have several opportunities to establish disability, and many people first apply online before contacting an attorney after a denial. Nick describes attorneys as translators of a complicated regulatory system. Their work is not limited to completing forms. They must build a record that connects the claimant's condition to the rules.
Nick's high-volume practice exposed the cost of manual case preparation
After graduating from law school in 2011, Nick moved from substitute teaching into Social Security disability practice. His firm handled a large number of cases, and he personally handled about 300 cases a year while driving roughly 40,000 miles to hearings around Arkansas and the region. He also drafted appellate work, which required detailed research into errors made by the Social Security Administration. Handling both responsibilities created pressure around malpractice, ethics, and whether he could advocate effectively for so many people. Nick says this experience pushed him toward technology. He was already a computer enthusiast who looked for ways to automate repetitive processes.
Medical evidence must be assembled like a puzzle across many visits
Nick uses Listing 1.04 as an example of why disability review is difficult. A claimant may need diagnostic imaging showing nerve compression, a physical examination showing limited range of motion, sensory or reflex loss, muscle loss, and a positive straight leg raise test. In a 2,000-page medical record, these findings may appear across separate visits rather than in one note. An attorney therefore has to piece together the record and show that the required factors occurred within the relevant period. Nick previously used Adobe and plugins to build word repositories that helped locate evidence. Those tools were better than manually searching every page, but the process remained slow.
Large language models handle medical synonyms that simple search misses
Nick says his earlier search system could flag terms connected to the regulations, much like an advanced version of Control-F. The problem was that medical records do not use one consistent phrase for a finding. Reduced range of motion might be described with terms such as "limited" or "reduced," along with many other clinical expressions. Regular expressions and word repositories could not reliably connect all of those variations. Large language models helped because they can distinguish synonyms and similar phrases that have the same meaning under the regulatory framework. After OpenAI released its new tools in November 2022, Nick applied the decision-tree logic he had used manually to prompts and mapping techniques for LexMed.
LexMed maps changing regulations to changing medical records
Nick describes ChartVision as a platform that will use retrieval-augmented generation to connect a database of the regulatory framework with medical evidence. The medical record changes as a claimant receives new treatment or diagnostic testing. Someone who did not meet a listing in January 2025 might later receive a test showing nerve root compression. The rules can change as well. Nick says the requirements for Listing 1.04A became harder, adding requirements involving severely compromised arms or the need for a walker. A system that connects the current rules to the current record can help attorneys adapt their case strategy when the framework changes.
Hearing Echo applies legal review to transcripts and testimony
LexMed started with hearing transcription because the full medical-analysis product required substantial security work and investment. Nick says the transcription system uses hearing audio and human transcripts as ground truth. It looks for patterns in automatic speech recognition, uses regular expressions to correct words, and labels speakers based on who appears in the hearing. The product then applies Nick's review logic to identify errors in expert testimony. This gives attorneys a written set of findings instead of requiring them to listen through long, unpleasant recordings and rewind them while searching for evidence that could support an appeal.
Vocational testimony can be checked against rules and job data
At the hearing stage, a vocational expert gives an opinion about whether a claimant can perform other work if they cannot return to their previous job. Nick says experts sometimes cite jobs that are no longer present in the national economy. LexMed checks this testimony against the relevant regulatory framework and a SQLite database containing job information. The system also uses an MCP server and functions to reconcile whether a cited job exists and whether the claimant could perform it. Nick gives the example of a claimant who cannot bend at the waist more than occasionally being matched to a job that requires frequent stooping. The system flags the conflict and explains it.
Automation could reduce the economic triage forced on attorneys
Social Security disability practices commonly work on contingency fees, so attorneys must decide how to spend limited time across many clients. Nick describes having several hearings in one day and deciding whether a case with an uncertain outcome deserves hours of additional work. That calculation can leave some claimants with less-developed records, even when the attorney believes they deserve benefits. LexMed is intended to remove some of this triage by producing a cheat sheet of red flags and evidence. Nick wants attorneys to be able to review every case more thoroughly, including cases involving conditions such as fibromyalgia that might otherwise receive less time than a case with obvious nerve compression.
