Practicums · For students and programs
The exam room, simulated.
A patient who converses. An evaluator who scores. Separately. OSCE-style clinical skills assessment capacity, without standardized-patient scheduling and cost.
Demo film in production
The ninety-second film is ready: paste its YouTube embed URL into src/data/practicumsData.js and it plays here.
Two intelligences, separated
The actor never grades the candidate.
The Patient AI
Briefed on one person: her history, her symptoms, her personality, her worries. It converses in real time, in her voice, at her level of health literacy, and it never volunteers what sits behind a disclosure gate. Ask a general question, get a general answer. Ask the discriminating question, get the fact.
The Evaluator AI
A separate intelligence that reads the full encounter afterward and scores it against the station’s explicit key features: what was asked, what was examined, what was reasoned, what was missed. The actor never grades the candidate; the roles are separated in software the way an OSCE separates them in the room.
The instructor, in charge
A three-panel review workspace: the transcript, a per-domain scorecard with instructor override on every score, and a chat where the instructor can question the evaluator about its own reasoning. Override is a permanent part of the design, not a temporary concession.
Try the miniature on the right: general questions get general answers, the discriminating question opens the gate, and a separate evaluator scores the transcript afterward, hurdles first.
Before you begin
A scripted miniature; the product is spoken, on video
Margaret Chen is a 74-year-old woman with three weeks of worsening breathlessness. Take a focused history. You have three questions.
Question 1 of 3: what do you ask?
Anatomy of an encounter
From the door instruction to the scorecard.
The door instruction
Who the patient is, what you are asked to do, and any constraint on how. Reading time is yours; nothing runs until you press Begin, exactly as on the real exam.
The room
The patient on video, a live transcript filling in as you both speak, an elapsed timer, a mic toggle, and a red End session button. Your camera matters: conduct is observed, and the patient reacts to it.
The conversation
Just talk. The patient answers what you ask, at their level of health literacy, and never volunteers what sits behind a gate. Their affect responds to your conduct turn by turn, and some changes, once triggered, do not go back.
The examination
You examine by saying what you are doing, out loud. When the patient recognizes the request, an overlay presents the finding: a video clip with synced auscultation, an image, or a report. Your mic is muted to the patient while it is up, so you can think aloud.
Investigations
Same mechanism: ask for the ECG, the bloods, the film, and read the result yourself, laid out like a real report. Every request is recorded before the finding returns: there is no way to look at a result invisibly, so ask for what you have a reason to ask for.
The note, then the scorecard
Stations with an oral-question equivalent ask for a written note: diagnosis and management plan, required, before any feedback. Then the evaluator reads everything and the scorecard is built.
Watch a full encounter
Demo film in production
The full-encounter film is ready: paste its YouTube embed URL into src/data/practicumsData.js and it plays here.
How scoring works
A good-feeling encounter can still be marked down. The report explains why, in the right order.
Key features, with explicit logic
A station is scored against the specific things a competent candidate must do on this presentation, each with explicit logic: all of, at least N of, any of. Partial credit inside a key feature does not exist; two of three named investigations is not two thirds, it is not satisfied.
Hurdles before numbers
Safety-critical features weigh 3, essential features 2, everything else 1. Missing any essential feature fails the station’s hurdle regardless of the percentage, and the report says so first: hurdles, then key features, then the number last. Showing a comfortable percentage above a missed safety item would teach the wrong lesson.
Abstention, never a guess
Where there is genuinely no evidence either way, the evaluator abstains, and abstained items are dropped from both sides of the fraction rather than counted as zero. An abstained item is never a miss against the candidate. Attempts below the engagement floor are stated plainly as too short to judge.
Formative first
By default a student sees coaching without the number: strengths, gaps, and the behavioural markers, which are observations rather than marks. The full scorecard is released when the instructor releases it.
Competency ratings run on a five-anchor scale, from unacceptable to above the expected level, and the reference point is a recent graduate accepted into postgraduate training, not a consultant. Acceptable is a good outcome, and the product says so.
For programs: building a station
A case is a curriculum object.
One document, whole station
A case is a single structured document: patient identity and demographics, avatar and voice, chief complaint and door instruction, the full history, social and family context, personality and behaviour, physical findings, and the key features it will be scored against.
Findings as media
Physical examinations and investigations are authored as the material the student will actually see: examination video clips with synced auscultation, images, laboratory panels laid out like real reports.
Blueprint-mapped
Cases carry blueprint metadata, so a station is a curriculum object: what it teaches, what it assesses, and where it sits in a program’s map.
A deliberate publishing gate
Cases move draft to published through a one-time gate. Clinical content authored with AI assistance waits for physician certification before students see it. That gate exists deliberately and is not shortcut.
Building a case
Demo film in production
The case-building film is ready: paste its YouTube embed URL into src/data/practicumsData.js and it plays here.
What we claim, and what we do not
Said plainly, because you will check.
Validation is the standard, and the study is not done
The evidence that the evaluator’s scoring agrees with human examiners is the standard this product is built to meet, and the validation study has not been run yet. We say that plainly rather than implying otherwise.
The instructor holds the pen
Every score carries per-domain instructor override, and the default release model is formative coaching first, numbers when the instructor releases them. The human stays in charge of consequence.
Certified before a student sees it
AI-assisted clinical content passes physician certification before it is served. The same liability rule that governs the rest of EdAI governs the simulation lab.
What it replaces, and what it does not
Practicums gives programs OSCE-style assessment capacity without standardized-patient scheduling and cost, for practice and formative evaluation. It is not a regulator, not an accreditor, and not an endorsement; those words belong to the authorities.
Encounters land in the institutional record.
Practicums integrates natively with EdAI MedSchool: stations map to the curriculum, and the competency evidence each encounter generates lands in the same record a school shows its accreditor.
Questions programs ask