At the Arizona State Urological Institute, the bottleneck had a familiar shape: a fax. Hundreds arrived each week. One employee had to sort them, find the referrals, and enter the information into the medical record. A document could take up to half an hour to handle. From arrival to a populated record, the wait could stretch to four to eight hours.
There is an entire software business hidden inside that interval. Leapfrog Technology found one by helping the institute deploy Addy, its referral-processing system. The interesting part is how little the basic transaction changed. A referring clinician could still send a fax. The receiving clinic got a different way to deal with it.
- Leapfrog sells product engineering, AI implementation, cloud services and extra engineering capacity.
- Healthcare is a recurring specialty, with customers including ASUI and Laudio.
- Its useful habit: define the workflow, test a bounded change, then decide how far to expand.
The queue is the clue
Addy uses Amazon Textract to read incoming documents and Amazon Bedrock models to extract details such as patient demographics, insurance identifiers and diagnosis codes. It then populates the clinic’s electronic medical record. Leapfrog’s product description also includes notifying staff for review. The destination matters as much as the extraction: useful information has to reach the place where someone can act on it.
In its ASUI case study, Leapfrog reports that handling time fell to two to five minutes per case, with more than 95 percent extraction accuracy and nearly 1,900 administrative hours freed annually. Those are vendor-reported results from one deployment. They describe a particular workflow, rather than a promise that every clinic will get the same improvement.
The first constraint here was manual intake capacity. More documents meant more work for the same person. Leapfrog changed the intake machinery while keeping the incoming channel. For a buyer looking for an AI project, that is a useful starting question: where does an ordinary task repeatedly stop moving?
A bridge built from Nepal
Leapfrog’s history begins well before the current enthusiasm for generative AI. In a 2018 interview, founder Himal Karmacharya described leaving Deerwalk in 2010 and starting Leapfrog with former Oracle colleagues. An early project was mobile banking software developed with Nepal’s Kumari Bank. The ambition combined international software work with building something in Nepal.
Today the company is headquartered in Seattle and has a substantial engineering presence in Kathmandu. Its work spans product discovery, interface design, web and mobile applications, data engineering, testing and cloud operations. A customer can commission a project, assemble a dedicated squad or add specialists to an existing team. These are different purchases: a team needs a different kind of direction from a narrowly defined implementation.
The talent model extends beyond hiring for current vacancies. Leapfrog offers paid fellowships with direct mentors and work on real projects. Its public GitHub organization shares coding guidelines and tools, including a Nepali date picker. Those details give the company’s learning culture a visible form: people can inspect some of the practices and software behind the pitch.
The product team that stayed
Laudio makes the dedicated-team model concrete. Its software helps frontline healthcare managers organize work and maintain relationships with staff. According to Leapfrog’s case study, the partnership began in September 2017 with discovery, design and development of a minimum viable product. That first product launched within three months.
The relationship then grew. Leapfrog describes an initial five-person discovery and design team becoming a 50-person group spanning application development, data, AI, cloud operations, design and product management. The work included integrating health-system data and maintaining the infrastructure. Keeping a product useful became a continuing assignment.
“It would have taken months to build the product team that we inherited overnight with Leapfrog.”Russ Richmond / CEO, Laudio
That helps explain Leapfrog’s place in a crowded market. Its pitch joins product judgment, delivery capacity and knowledge of healthcare systems. Buyers can compare it with firms such as Globant or Endava, or build their own team. The relevant comparison is whether a partner can understand the problem and keep improving the software after launch.
Give the model a smaller job
A newer project offers a precise example of that judgment. Binq, a UK fintech, wanted business owners to understand their credit scores. Leapfrog’s September 2026 case study describes a five-month journey from kickoff to a subscription product with paying customers.
The architecture gives the language model a defined role. Experian supplies the core score. Binq’s code calculates the surrounding financial measures. Claude receives the finished values and explains them in plain English. The model cannot change those calculations. A conversational assistant adds questions and answers, with guardrails designed to keep responses descriptive rather than advisory.
The lesson is portable: decide which part of the job needs generated language, and which part needs a fixed answer. That boundary is especially consequential when customers use an explanation to think about their finances.
External credentials add another piece of evidence. AWS listed Leapfrog for DevOps Competency in May 2026, Agentic AI Consulting Services in July, and Healthcare Competency in September. These designations validate particular areas of practice. Buyers still need to examine the proposed team and the work it will do for their own organization.
Buy the next decision
Leapfrog packages some discovery work into short engagements. Its healthcare AI proof-of-concept listing specifies four to eight weeks, defined success measures, real integrations and a roadmap. Implementation pricing comes through a private offer based on requirements. A separate healthcare consultation is advertised as free.
A prototype also creates obligations. Leapfrog’s production offering includes evaluation, monitoring, security, human review and incident procedures. For buyers, the practical implication is to budget for running the system as well as building it. The approach needs accessible data, an owner for the workflow and people who can judge exceptions. Standard needs with a good existing solution may call for a simpler purchase.
What readers can copy is the sequence: locate the queue, establish the baseline, test a bounded intervention and measure what happens. At ASUI, the telling detail was a person spending half an hour on a document. An ambitious software project can begin by taking that person’s afternoon seriously.