Introducing CLARA

Turn any clinical PDF into iterative learning questions in minutes

What is CLARA?

CLARA a Clinical Learning And Reasoning Assistant, a free web tool, built by LITFL and Medmastery, that turns any clinical PDF into examination-style practice questions.

CLARA produces fully editable board and Fellowship style clinical vignette questions with one‑best‑answer, five options and evidence‑grounded explanation. This first iteration has been designed for college exams, fellowship boards, in‑training exams, and CME / CPD self‑assessment.

Educators upload, review, publish, and share. Learners click a link, take the quiz in the browser, and get instant feedback. No login is required for learners.

Why build CLARA?

We all know what a good clinical practice question looks like.

It needs a real patient vignette with clinical history, vitals, labs etc. The question must force a clinical decision, not simply test recall. The distractors must be wrong for a specific clinical reason; not obviously wrong and not impossibly obscure. The explanation must say exactly why each option is right or wrong for the clinical scenario. And the key learning points should be highlighted…

But this takes a lot of time.

Faculty can easily spend up to an hour on a single high‑quality MCQ. Even then, the literature shows that 37% of faculty‑written items contain at least one measurable flaw such as implausible distractors or the correct answer being the longest option topping the list.

Creating plausible yet incorrect distractors is more challenging for question writers than any other element of the item.

National Board of Medical Examiners

Writing a great clinical question is not about clinical knowledge. It is about question construction, a skill most clinician‑educators have never been formally taught.

We have both felt the frustration trying to write meaningful and challenging question to test examination candidates or create iterative self-learning challenges. Question sources such as the new AHA, ACC, ESC or ACCP guidelines, the latest trials, or recently read textbook chapter sitting idle as a PDF on your desktop can now be used to design stimulating and challenging questions in minutes.

Why create questions?

You may read a guideline three times you feel confident in your level of understanding. Yet when you sit your exams totally fail questions based on that same guideline

Reading and answering questions are completely different skills

Reading is recognition. Question‑answering is retrieval and reasoning under pressure. They use different cognitive muscles, and only one of them is what shows up on exam day.

Answering clinical questions on the material you have just read forces two things at the same time. It tests your comprehension under the same conditions the exam will. And it builds the pattern recognition such as “why is this distractor wrong for this specific patient?“. This separates higher scorers from those who simply know the content. Tearing apart the distractors is where the highest‑yield learning happens.

That is why almost every Medmastery course and every evidence‑based LITFL deep‑dive ends with questions. We built CLARA so the writing of those questions stops being the bottleneck.

How CLARA works for the educator

1. Sign up free. Account in seconds, email confirmation.
2. Upload a PDF. Drag and drop, up to 15 MB. CLARA extracts the citation, authors, journal, year, DOI, PMID, PMCID so the source is always cited correctly back to the original reference.
3. Pick how many questions to generate. Anywhere from 3 to 30. CLARA reads the document and writes each item with a clinical vignette, a closed lead‑in, five same‑category options, an evidence‑grounded explanation, two to four pearls, and the guideline section reference plus COR / LOE evidence level.
4. Review and edit. Every wording, option, classification (organ system, physician task, discipline) is editable in a side‑by‑side editor. Add or delete questions. Generate more on demand. Click Publish and you get a permanent share link.

Behind that share link sits a feedback dashboard including thumbs up / thumbs down on each question, written comments, take counts, certainty breakdowns. You see exactly which questions are landing.

How CLARA works for the learner

1. Click the link your educator shares. No account, no signup.
2. Answer one question at a time. Mark how certain you are. This separates I knew it from I guessed right, a most important distinction in educational learning.
3. Get immediate feedback. Combination of correctness with certainty (e.g. “You were certain, but this was actually wrong. Confident mistakes are the most important ones to address.”).
4. Read the explanation. Evidence level (e.g. COR 1, LOE A), and a direct citation back to the source paper or guideline section.
5. Vote and comment. Provide feedback to help the educator refine the question.
6. Repeat. Retry the questions you got wrong.

Are the questions any good?

Most clinicians have asked ChatGPT, Claude or Gemini for ten USMLE‑style questions and watched them hallucinate, make up references, invent a trial that does not exist or quote a cutoff that is wrong. This can be very frustrating. So here is what we did differently in CLARA.

• USMLE‑faithful format. Every question follows the one‑best‑answer structure: a 3–6 sentence vignette, a closed lead‑in, five same‑category options.
• Source‑faithful generation. A two‑phase pipeline first extracts a structured reference table from the source – every threshold, contraindication, classification boundary – and feeds it to the question writer as the single source of truth. No hallucinated trial names. No invented cutoffs.
• Clinical validity self‑check. Severity tiers must match (the right answer matches the patient’s actual severity, not an adjacent one). Every vital, lab, score, and age is checked against guideline thresholds with the correct relational operator.
• Distractor quality. Distractors are tempting but defensibly wrong. The kind of mistakes a real learner would make, not absurd options that give the answer away.
• Answer‑position balanced. A through E each appears with no learnable position bias.
• Topic spoiler protection. Topic labels never give away the answer and only appear after the learner reveals the option.
• Cited and traceable. Every question links back to the section it was drawn from, with DOI, PMID, PMCID, and section reference. No “trust me, it’s in the literature.”

---

Three ways to use CLARA

We would love your feedback

This is the first iteration of CLARA, and we would really appreciate your feedback. CLARA is live, but very much in its infancy. What it grows into depends on the people using it.

Inside every quiz you will see a thumbs‑up / thumbs‑down on each question and a free‑text comment box. Please use them. That feedback flows straight back to the question author and to us. It is how the next iteration gets built.

---

Related reading and resources

• Haladyna TM, Downing SM, Rodriguez MC. A Review of Multiple-Choice Item-Writing Guidelines for Classroom Assessment. Applied Measurement in Education, 2002; 15(3): 309–333.
• Roediger HL 3rd, Karpicke JD. The Power of Testing Memory: Basic Research and Implications for Educational Practice. Perspect Psychol Sci. 2006 Sep;1(3):181-210. Why retrieval practice beats re‑reading. The reason we put questions at the end of every Medmastery course and every evidence‑based LITFL deep‑dive.
• Wiesbauer F. Teaching Masterclass. Medmastery. How to design effective teaching, including how to write assessment items that actually measure clinical reasoning.
• Wiesbauer F. ChatGPT Essentials for Clinicians. Medmastery
• NBME Item‑Writing Guide – the gold‑standard reference for one‑best‑answer item construction. Free PDF from the National Board of Medical Examiners. Read it once and your editing of CLARA’s drafts will get sharper.