.jpg)
Question banks are at the heart of how every medical student studies, but their dashboards have always looked backward. Most Qbank providers tell you how many questions you've done, percentage correct, and scores by subject. The AMBOSS Knowledge Profile looks forward instead. With every question you answer, it updates a live picture of what you've truly mastered, and uses it to guide your next session, show your exam readiness, and track your progress in the long run.
Why we built the Knowledge Profile
Learner behavior and the limits of conventional analytics pointed to the same set of problems:
- Traditional performance metrics are unstable. Percent correct is a shaky measure. It reflects which questions you happened to get, not how much knowledge you've acquired. Imagine two learners who both score 70%. One answered mostly easy questions; the other took on the hardest ones. Same number, very different readiness
- Weaknesses are surfaced too broadly to act on. Most question banks flag weak areas at the level of a subject or a body system. That identifies a region of difficulty without answering the question learners actually ask, which is what to study next.
- Your data does not carry forward. When you move between exams, most conventional dashboards start from zero, meaning you have no idea how much of the new exam content you’ve already covered. Your knowledge doesn’t reset, so why should your progress?
- Score prediction has traditionally required a mock exam. Learners could only benchmark themselves by setting aside time for a practice exam. Between those sittings, they had no reliable read on whether their studying was moving them forward.
- Percent correct doesn’t reflect your progress over time. Since it shows the average from your entire history, raw percent correct doesn’t always reflect where you’re at right now, and recent improvement takes a while to appear.
What makes our approach unique
The Knowledge Profile applies the principles of precision medical education: instead of relying on a few isolated, high-stakes exams, it uses continuous learner data to deliver personalized feedback and timely support.
Technically, the Knowledge Profile is a knowledge tracing system based on item response theory, specifically a Rasch model. Every question in the AMBOSS question bank is tagged with the skills it tests, across dimensions such as discipline, body system, clinical task, and medical topic. For each question, the model estimates how likely you are to answer it correctly, based on two things: the learner's ability in the relevant skills and the question's difficulty. Instead of one overall score, it maintains hundreds of separate ability estimates per learner.
This makes the profile both precise and lasting. First, the estimates are granular enough to act on: they describe proficiency in individual topics, such as pneumonia or myocardial infarction, and in cross-cutting competencies, such as formulating a diagnosis. Second, because the model is organized around medical knowledge rather than a single exam, it builds a persistent picture of what a learner knows, one that keeps growing across their entire education.
How the Knowledge Profile fits into the learning journey
- In the preclinical years
Early question bank practice is often discouraging, because raw scores at this stage say more about a learner's stage than their progress. The Knowledge Profile instead identifies which specific topics are weakest and tracks whether proficiency is improving. In the short term, it creates adaptive Qbank sessions to help students close knowledge gaps immediately. In the long term, every question answered contributes evidence about the user’s strengths and weaknesses, which remains useful even years later.
- During exam preparation
As an exam approaches, the model translates ordinary studying into a readiness check. Adaptive Qbank Sessions target the exam-relevant gaps most likely to affect performance, while Equated Percent Correct, Probability of Passing, and the Exam Score Estimate show where a learner stands without requiring a practice exam.
- Across exams and into clinical training
Because accumulated evidence carries forward, when learners transition between study objectives, for example, from Step 1 to shelf, or from Step 2 CK to specialty board exams, they begin with an established picture of their strengths and gaps rather than starting from zero.
How we validated it
How close are AMBOSS’s estimates to students’ real exam outcomes? Very!
AMBOSS readiness metrics and adaptive Qbank sessions rely on the accuracy of the underlying model. Exam score estimates were compared against self-reported USMLE Step 2 CK scores from 7,244 learners. The estimates achieved a mean absolute error of 8.52 score points, reducing prediction error by approximately 23% relative to a baseline that predicts the dataset average for every learner.
That figure is best read against the theoretical limit. Since any test-taker would score differently at different sittings, the standard error of the estimate of 8 points reported in the USMLE Score Interpretation Guidelines implies a minimum achievable mean absolute error of roughly 6.4 points. At 8.52 points, AMBOSS’s model is approaching the practical limit of predictive accuracy. It is unique because it gets these insights from routine question bank activity rather than dedicated mock exams.
The full methodology, the validation analysis, and a discussion of the model's current limits are set out in a white paper authored by AMBOSS Senior Data Scientist Christina Chang.
How the Knowledge Profile expands the AMBOSS experience
The Knowledge Profile is not a separate tool; it’s a new layer underneath features learners already use. Adaptive Qbank Sessions, which launched in April 2026 and have since been used by more than 48,000 students, draw on it to sequence questions. The readiness metrics in Analysis are derived from it. Performance over time charts it.
The Knowledge Profile is tailored to the learner's study objective, which they set in their account. Equated Percent Correct and Adaptive Qbank Sessions are supported for every study objective. Probability of Passing, benchmarked against the standardized exam, is available for USMLE Step 1 and Step 2 CK and for COMLEX Level 1 and Level 2. The Exam Score Estimate is available for the exams that report a score, namely Step 2 CK and COMLEX Level 2. Metrics appear once a learner has answered 40 questions, and all of it is included in every AMBOSS membership at no additional cost.
Work continues in three directions. The Knowledge Profile connects to AI Mode Learning, so that a learner's ability, question performance, and exam readiness all factor into the guidance they receive and the study plans generated for them. Future versions will incorporate evidence beyond question responses, including AI interactions and other study activity such as Anki. For educators and institutions, the model creates a foundation for programmatic assessment at scale, aggregating many low-stakes observations into a continuously evolving picture of learner development. This also serves the longer-term goal of mapping learner knowledge directly onto a school's curriculum to identify struggling learners earlier and evaluate curriculum effectiveness.
A learner's Knowledge Profile is specific to them and is built solely from their own activity. AMBOSS follows strict privacy and security standards, including EU data processing requirements, to protect learner information.
For support or additional questions, contact hello@amboss.com.
