The Missing Piece in Medical Education Is Practice at Scale

01 /The Concept

One of the most important – and hardest to scale – components of medical education is the patient encounter: the interview and physical examination through which students learn to identify problems and form appropriate recommendations.

As it turns out, this is also one of the most expensive parts of running a medical school, costing institutions hundreds of thousands of dollars a year.

The reason is structural. Students can only practice these encounters in supervised classroom settings. That means hiring trained actors to play patients, organizing the sessions, and paying faculty to oversee each one – individually, one student at a time. Lectures and seminars can be scaled by packing hundreds of students into a hall or moving online. Patient encounter practice cannot. This fundamentally limits how quickly and cheaply medical schools can train large numbers of doctors.

Soma Lab addresses this directly by providing medical schools with AI patients that students can interact with just as they would with real people.

These AI patients can present with different conditions and have different personalities – affecting how they communicate, how forthcoming they are with information, and how cooperative they seem. The student listens, questions, probes, and must ultimately arrive at the right diagnosis and recommendation.

The platform evaluates more than just communication style. At the end of each session, students receive instant feedback covering the quality and structure of their questions, the diagnostic conclusion they reached, and the appropriateness of their recommendations.

Instructors, meanwhile, can access session reports alongside the platform's feedback – giving them a clear basis for planning individual work with each student.

The platform's potential client base extends beyond medical schools. Telehealth companies and social care organizations can use it for ongoing staff development, as well as for assessing candidates during hiring – testing their remote patient communication skills before they're deployed in the field.

The platform found immediate demand: Soma Lab secured 30 pilot programs in just three weeks of outreach. The startup is currently in Y Combinator, from which it received its first $500K.

02 /Why It Matters

A [recent review](/review/chtoby-zarabotat-mnogo-deneg-uchit-nuzhno-prostym-veshham) covered the shortage of junior medical staff – medical assistants and pharmacy technician trainees in particular.

The week before that, [another piece](/review/glavnoe-vybrat-pravilnyj-rynok-a-ideju-mozhno-najti) looked at the licensed nursing shortage – which the US Chamber of Commerce has called a national crisis.

Today we arrive at the doctor shortage. The president of the American Medical Association recently flagged it as a national problem as well. Various estimates project a shortfall of up to 100,000 physicians in the US – the majority of them primary care doctors, whose core function is talking to patients: understanding their problems, deciding what to treat in-office, and knowing when to refer.

Scaling the output of medical schools is the obvious starting point for addressing this. But scaling the practical component of training – the part that Soma Lab now makes possible with AI patients – has been the missing piece.

Another Y Combinator company puts this shortage in a different light: Asha Health ([covered previously](/review/razmer-vygodnee-chem-slozhnost)) is building AI bots to handle routine consultations for patients with chronic conditions. As it turns out, 76% of all doctor visits are made by people with chronic illnesses, many of whom can be helped with straightforward, timely guidance that AI is now capable of providing. If that model takes hold, it could meaningfully reduce the load on existing physicians and partially offset the shortage.

03 /Opportunities

The broader direction here is embedding AI characters into the training of professionals whose work critically depends on real-world interaction with people.

For most of these disciplines, training is currently heavy on theory, and practitioners only develop genuine competence once they're on the job – which substantially delays the point at which they reach the necessary level of skill.

Crucially, these AI characters shouldn't be generic or averaged-out. They need to have different problems, different personalities, and different communication styles – just like real people.

Hyperbound ([related review](/review/za-takoe-obuchenie-kompanii-tochno-zaplatjat)), which came out of Y Combinator and subsequently raised $1.5M, built exactly this for sales training: AI personas with distinct roles and distinct characters – "aggressive," "friendly," "overconfident" – so reps can practice handling a genuinely diverse range of prospects.

What other professions critically require practiced communication skills across a wide range of human personalities and situations? Law enforcement comes to mind immediately. But there are certainly less obvious fields where the same need applies.