Clinical-interview training
Practice taking a focused history. The AI patient minimizes symptoms, contradicts themselves, or volunteers irrelevant detail, the way real patients do.
Coming soon
Build LMS-ready diagnostic-interview training with Claude or ChatGPT.
Build question-asking training in plain language. Learners interview an AI character who's holding a hidden problem. The transcript is graded on which questions were asked, in what order, and whether the learner reached the correct diagnosis.
What is it?
What is a diagnostic-interview activity?
An interactive training activity where the learner interviews an AI character who is holding a hidden problem, a customer with a vague complaint, a patient with subtle symptoms, an engineer with a flaky bug. The learner must ask the right questions to surface what's really going on, and the transcript is graded on question quality, sequence, and final diagnosis.
-
AI character who holds a hidden problem and reveals information only when asked the right questions -
Graded on question quality, not just whether the learner reached the right answer -
Different runs use different hidden problems so memorizing isn't a strategy -
Per-question feedback explaining what each question surfaced and what it missed -
Exports as a SCORM 1.2 package for Cornerstone, Moodle, Canvas, TalentLMS, and every LMS -
Same bundle runs standalone in any browser for self-study without an LMS
How it works
How to build a diagnostic interview
in four steps.
From a plain-English description of the hidden problem and the diagnostic rubric, to a SCORM-ready interview simulation.
01
Describe the character and the hidden problem
Tell your coding agent who the character is and what's really going on. The character only volunteers information that's appropriate for what's been asked.
02
Define the diagnostic rubric
Name the questions a skilled interviewer would ask, in roughly the right sequence. The AI grades each turn against this rubric, with credit for variation on the same intent.
03
Refine the character in chat
"Make the customer minimize the symptom on first ask." "Add a red herring that points toward a wrong diagnosis." Iterate by talking, not by editing dialogue trees.
04
Export as a SCORM 1.2 package
One command produces a SCORM 1.2 zip. Upload to any SCORM LMS. Diagnostic-quality scores post back to the gradebook automatically.
Use cases by role
Diagnostic-interview training
for every job that depends on good questions.
The same toolchain powers clinical-interview training, sales-discovery practice, technical-support troubleshooting, root-cause analysis, and qualitative-research interviewing.
Discovery-call practice
Train reps to ask questions that surface real business pain instead of features-and-functions checklists. The AI buyer rewards open-ended questions and stonewalls leading ones.
Troubleshooting interviews
Practice the diagnostic conversation between support engineer and customer. The AI customer doesn't know the technical vocabulary, so the engineer has to ask the right questions.
Bug-triage interviews
Practice extracting reproducible bug details from a non-technical reporter. Graded on which questions narrow the search space and which waste time.
Qualitative-research interviewing
Train UX researchers and journalists to ask non-leading questions that uncover real user behavior, not the answers they were hoping for.
Incident-investigation interviews
Practice the post-incident interview. The AI subject is defensive, vague, or over-eager to assign blame. The learner is graded on extracting facts without contamination.
Question-graded interviewing
Every turn graded on which questions were asked and what they surfaced.
When the interview ends, the AI scores each turn against the diagnostic rubric. Credit goes to the right kinds of questions in the right sequence, not just to a learner who stumbles into the right diagnosis through luck.
Update the rubric, hidden-problem pool, or character behavior any time. Re-export the SCORM package and the next attempt grades against the new rules.
Per-question rubric scoring
Each turn is graded against your rubric, did the learner ask an open-ended question, was the sequencing logical, did they follow up on the implicit signal. Evidence quoted from the transcript.
Hidden-problem mechanics
The AI character holds information that is only revealed when the learner asks for it. Skipping a question costs the learner that information for the rest of the interview.
Randomized hidden problems
Different runs use different underlying diagnoses drawn from a pool. Learners can't memorize the right answer because the right answer changes.
Session complete
Your Debrief
9/10
★★★★★★★★★☆
Total 9 / 10
Asked an open-ended chief-complaint question before narrowing to closed yes/no items.
Followed up on the patient's minimization, instead of accepting 'it's nothing' at face value.
Reached the correct diagnosis with evidence traceable to specific learner questions.
Plain-English authoring
Author diagnostic interviews in plain English. No dialogue tree required.
Change the hidden problem, soften the character, add a red herring, or rewrite the rubric, just describe the change in your own words. Your coding agent rebuilds the character and re-exports the SCORM package without you ever editing a dialogue tree.
When you're happy, export as a SCORM 1.2 package for your LMS, or use the same HTML bundle standalone.
Make the patient minimize chest pain on the first ask, only reveal it on a follow-up.
Updated. Patient now downplays chest pain unless the learner asks an open-ended 'tell me more' question.
Add a red herring that points toward GERD instead of cardiac.
Added red herring
recent-spicy-meal. Wrong-diagnosis path now scores lower if not investigated further.Export as SCORM.
✓ diagnostic-interviews.zip ready (SCORM 1.2, 14 KB)
Interview Library
Start from real diagnostic
conversations.
Paste any prompt into your coding agent to get a complete diagnostic-interview simulation as a single self-contained HTML file. Adjust the character, hidden problem, or rubric in plain language.
Clinical
Atypical chest pain
A patient with non-classic chest pain who minimizes the symptom and volunteers a GERD-flavored red herring. Tests follow-up question quality.
Prompt
Using /diagnostic-interviews, build a 14-turn chest-pain interview. Score on open-ended history, red-herring resistance, and cardiac vs GI differential.
Sales
Buyer with hidden urgency
An enterprise buyer who has a real deadline but doesn't volunteer it. Practice the discovery questions that surface real urgency vs polite interest.
Prompt
Using /diagnostic-interviews, build a 10-turn enterprise discovery interview. Score on uncovering hidden urgency, decision-process mapping, and economic-buyer ID.
Support
Flaky-bug customer call
A non-technical customer reporting a 'sometimes broken' bug. Practice extracting reproducible details without overwhelming the customer.
Prompt
Using /diagnostic-interviews, build a 12-turn flaky-bug support call. Score on reproduction-step extraction, frequency capture, and non-technical clarity.
Engineering
Production-incident eyewitness
An on-call engineer interviews a teammate who first noticed an outage. Practice surfacing timeline and signals without leading the witness.
Prompt
Using /diagnostic-interviews, build an 8-turn incident-eyewitness interview. Score on timeline accuracy, signal capture, and non-leading questions.
Research
User-research session
A UX researcher interviews a participant who keeps trying to be helpful by guessing what the researcher wants to hear.
Prompt
Using /diagnostic-interviews, build a 10-turn user-research interview with a please-the-researcher participant. Score on non-leading questions and observed-vs-desired behavior.
Investigations
Post-incident interview
An investigator interviews a defensive employee after a near-miss safety incident. Practice extracting facts without contaminating memory or assigning blame.
Prompt
Using /diagnostic-interviews, build a 12-turn post-incident interview. Score on factual extraction, non-leading questions, and blame neutrality.
SCORM & LMS
SCORM 1.2 diagnostic-interview activities for any LMS.
Every diagnostic-interview activity exports as a standards-compliant SCORM 1.2 package. Upload the zip, assign it like any other course activity, and diagnostic-quality scores flow back to the gradebook automatically.
-
Single SCORM 1.2 zip, upload to your LMS, no integration work, API keys, or custom JavaScript -
Completion, score, and time-on-task post back via cmi.core.score.rawandcmi.core.lesson_status -
Tested with Cornerstone OnDemand, Moodle, Canvas LMS, TalentLMS, Docebo, Brightspace, Absorb, 360Learning, SuccessFactors, and Workday Learning -
Same self-contained HTML bundle runs standalone on a public page, an intranet, or as an embedded iframe -
Hidden-problem pool randomization makes the activity replayable and prevents memorization -
Grading runs inside the bundle, no middleware, no data warehouse, no analytics SDK required
FAQ
Frequently asked questions about
diagnostic interviews.
Authoring, hidden-problem mechanics, LMS compatibility, character realism, model choice, answered directly.
A diagnostic-interview activity is a training simulation where the learner interviews an AI character who's holding a hidden problem. The learner must ask the right questions to surface what's going on, and the transcript is graded on question quality, sequence, and final diagnosis.
A role-play is an open conversation graded against rubric objectives, the character could be anyone with any goal. A diagnostic interview is specifically about question-asking, the character is holding a hidden problem and the learner is graded on how efficiently they uncover it.
Yes. You can define a pool of underlying problems, and each learner attempt draws a different one. This makes the activity replayable and prevents learners from memorizing the right answer.
Each turn is graded against the rubric of questions a skilled interviewer would ask. The learner gets credit for question types and signals followed up on. The final score combines question quality with whether they reached the correct diagnosis.
Yes. Every activity exports as a SCORM 1.2 package, which works in Cornerstone, Moodle, Canvas, TalentLMS, Docebo, Brightspace, and every other SCORM-compliant LMS. Scores post back to the gradebook automatically.
Yes, that's the point. You can configure the character to minimize symptoms, contradict themselves, volunteer red herrings, or get defensive. The interview gets harder the more realistic the character's behavior.
The bundles run on whichever frontier model you configure, Claude, GPT, Gemini, or a self-hosted model. The model is swappable; the rubric, hidden-problem definitions, and grading logic are model-agnostic.
Diagnostic Interviews is in active development. It will install via your coding agent with a single command, the same way the existing role-play skill works today. Add your email below and we'll let you know the day it ships.