How Physician Founders Build Clinical AI Tools

From the Clinic to the Production Codebase

As frontier AI models become increasingly capable, physicians are no longer just sitting on advisory boards for healthcare tech companies—they are actively building the software themselves. In a recent Offcall webinar, Dr. Tom Kelly, a vascular surgeon and the CEO of Heidi Health, revealed exactly how he uses AI to contribute directly to his company's production software.

Before models like Claude 3.5 Sonnet (Opus 4.5 era), AI was primarily useful for building small, lovable prototypes. But working inside a massive production database with thousands of files and dozens of engineers was out of reach for non-developers. That changed recently.

Even while running a rapidly growing startup, Dr. Kelly remains a highly active, hands-on builder. "I did like 120 PRs [pull requests] that got into production over like January, February," he shared. His contributions ranged from minor UI styling fixes to building out entirely new, complex Continuous Professional Development (CPD) tracking features from scratch before handing them off to engineers for backend refinement.

Why Clinical Expertise is a Superpower in Tech

Historically, doctors with great ideas had to rely entirely on engineering teams to bring their visions to life. This often resulted in a frustrating game of telephone, where clinical nuances were lost in translation.

Dr. Kelly pointed to Heidi's medical evidence grading feature as the perfect example of this friction. "Me trying to explain to like an engineering team the concepts of [Cochran review scoring systems] is so hard," Dr. Kelly noted. "It would take me to write like whole pages."

Vibe coding completely changes that dynamic. Because Dr. Kelly intuitively understands medical literature, journal impact scores, and clinical reliability, he could use AI to build Heidi's evidence-grading feature himself. By bypassing the traditional layers of designers and developers for the initial build, clinician-builders can iterate rapidly and build exactly what they know their peers need.

Overcoming the Technical Gap with Cursor

While Dr. Kelly does have an undergraduate background in computer science, he stressed that he is not a professional developer. To bridge the gap, he utilizes tools like Cursor, an AI-powered code editor with a built-in browser.

This setup allows for immediate, visual iteration. "You can basically, as you build it, once it shows you what it's done... you're like 'oh god this looks horrible' and you can actually click on the area and say 'I really was imagining like a tabulated view here,'" he explained.

This approach has even transformed how Heidi operates as a company. Instead of writing long product requirement documents, the team now communicates through code. Dr. Kelly will vibe code a feature, record a video explaining his logic, and push the code. The design team then iterates on his code to make it beautiful, and the engineering team refines the backend architecture for scale.

Advice for Aspiring Physician Builders

For doctors sitting on a great idea but lacking a clear path forward, Dr. Kelly offered highly actionable advice: focus on word of mouth.

"Find the early community that's going to like your tool, give you feedback, and then share it with their friends," he advised. Early in his journey, he reached out to a Facebook group of tech-curious doctors, offering to visit their clinics and show them a prototype. He found a small group of 10 to 20 physicians willing to test his ideas. Once they started asking how to sign up and pay for it, he knew he had a viable product.

The tipping point for leaving clinical practice? Obsession. "I found myself just spending all my time dwelling and thinking about like, 'oh man I want to add this'... I don't want to have to go to the next toe amputation," he joked.

The Physician Vibe Coding Showcase

The Offcall community is already acting on this builder mentality. The webinar closed with a rapid-fire showcase of impressive tools actively being built by practicing doctors:

• Audio Scholar: Built by nephrologist Dr. Nikhil Shah, this tool solves the problem of having no time to read journals. It allows users to choose a specialty and uses AI to turn the week's notable literature into a concise 10-minute podcast, complete with direct links to the original papers.
• Physician Builders: Created by emergency medicine physician Dr. Mike Logan, this is a healthcare sandbox. Clinicians can pitch ideas, match with engineers, build MVPs using synthetic patients and CMS payment models, and collaborate with a global network of 10,000 clinicians to upvote and iterate.
• Primary AI: A clinical decision support tool developed by family physician Dr. Rashad Usmani. It features verified citations, a localization feature for regional guidelines and antibiotic resistance patterns, and upcoming CME tracking capabilities.
• Claire: An AI triage assistant built by emergency physician Dr. Josh Barron. Claire manages informal, on-call medical requests by gathering clinical information, identifying urgency, and delivering decision-ready summaries to the physician.
• Whetstone Health: Designed by Dr. Erin Reinke, this platform leverages AI scribes to analyze patient encounters against over 30 evidence-based communication frameworks. It provides insights into strengths, highlights areas for improvement, and offers AI roleplay exercises for targeted communication training and CME credit.