criticalaccess.aiAI for rural hospitals

Prototype · Education only

An AI patient for clinical reasoning practice.

Health professions students need realistic practice interviewing patients and reasoning through cases, and instructors need a way to review it. This prototype gives a student an AI patient to interview, and feedback on the clinical note they write.

PythonFlaskPostgreSQLOpenAI APIBootstrap

01 · At a glance

The question, the data, and the terms.

A clinical reasoning trainer for health professions students. It is a prototype, and it is not hosted anywhere.

The question
Can students get realistic practice at interviewing a patient and reasoning through a case, in a form an instructor can review?
Who it’s for
Health professions students and their instructors.
Data
No outside datasets. GPT-4o generates the cases, and DALL-E 3 generates the patient portraits. Sessions, messages, and notes are stored in a PostgreSQL database for instructor review.
Status
A prototype. The repository, named Clinical_Tester, describes it as an “AI Clinical Simulation MVP.”
License
No license is stated in the repository.

02 · What it does

One practice session, start to finish.

The student works through one case, then gets feedback on the note.

  1. The tool writes a case

    It generates a patient case with hidden details the student has to find.

  2. The student interviews the AI patient

    The patient answers in conversation.

  3. The student orders a workup

    Vitals, an exam, labs, imaging, or a neuro exam.

  4. The student submits an assessment

    They write up what they think is going on.

  5. The AI gives feedback

    It covers what was caught, what was missed, and how the reasoning held up. Instructors can review the session afterward.

03 · What it found

Nothing yet. It hasn’t been validated.

The project states no findings and no validation results. It shows that a session like this can be built. It does not show that the cases are accurate or that students learn more from it.

What it takes to runNot a hosted tool

  • An OpenAI API key, for the cases, the patient, and the feedback.
  • A PostgreSQL database, for sessions, messages, and notes.

04 · Limits

For the classroom, not the bedside.

The project’s own documentation says: “Built for education only. It is not for clinical decision-making.”

Limits
  • It is for education only, and not for clinical decision-making.
  • It needs an OpenAI API key and a PostgreSQL database, so it is not a hosted tool.
  • The case content is AI-generated, and its accuracy is not stated.
  • The documentation lists no other limitations, and no validation results.

Teaching a class? Read the code first.

The prototype is on GitHub. If you try it with students, have an instructor review the cases it writes.