Our research platform
Students use qVault to revise. Researchers use it to study how learning happens. Built at King's College London, it brings teaching resources, formative assessment and education research into the same platform.
Run your study on qVault, end to end
Study learning inside the revision students already do. Agree the question with us, set up the study in qVault and follow the responses over time, with support through analysis and publication.
Design
Agree the research question, comparison and outcomes. We help configure the study, including randomisation where the design needs it.
Ethics
Agree the approvals and participant information for your study. We help prepare the documents and work through your institution's requirements.
Inform
Set out the study clearly for students, including how to take part, their choices and who to contact.
Run
Your conditions run inside the revision students already do: questions, explanations, flashcards, notes or Lenny.
Measure
Each answer records the result, how sure the student was before feedback, the time taken and what they did next. Students meet questions again over weeks, so you can follow how their answers change across a term.
Analyse
Agree the dataset and analyses in advance. Access follows the study's approvals, and reports combine small groups to protect students.
Publish
Publish what you find, subject to ethics approval. We work alongside you on the write up, from the first draft to peer review.
Some starting points
- How explanation design changes learning
- How AI written and human written questions perform with real students
- How confidence relates to later performance
- How students use AI tutoring, and what they ask
- Where misconceptions sit in a cohort, and whether teaching moves them
- Comparisons across modules, years or sites
Or bring your own question. Anything students do in qVault can be studied.
From revision to paper
Revision produces records of answers, study activity and use of the learning tools.
Names and account details stay behind. Only anonymised records reach the research team.
Your approved study follows learning across these records. Results are reported for groups.
We work with you to analyse the findings and prepare them for publication.
Every result, at the level you need
Choose a level and a lens to explore your teaching, from the performance of one question to patterns across a programme.
What students get right first time
Shows students' understanding before feedback, so you can see which parts of your teaching they grasped.
What students fix on a second try
Shows what improves with practice and what keeps slipping, so you know where practice alone is not enough.
Which wrong answer pulls students in
Names the misconception behind the most chosen wrong answer, so you can address it directly.
Where students slow down
Flags where students take far longer than usual, often a sign of unclear wording or a hard step in reasoning.
Where students are wrong while feeling sure
Finds the gaps students do not know they have, which a mark alone never shows.
When students revise
Shows when your students practise across the week and the term, so you can time practice before teaching moves on.
Published papers
Behavioral Sciences
Behavioral Sciences 15(9):1284, 2025
Medical students rated Lenny higher for usability, clarity and engagement in a randomised comparison with their usual study resources.
- Ease of use, satisfaction, clarity, perceived information quality and engagement were rated higher with Lenny in both study arms.
- Students valued quick answers to focused questions. They also wanted references, more support for reasoning and a closer fit to their teaching.
- Scores on the single best answer tasks did not differ significantly. Measuring experience alongside performance helped distinguish what students valued from what the tasks demonstrated.
- Focus group participants asked for questions, flashcards and diagrams from their own lecture slides, giving a practical brief for future study tools.
Scores in each task
Average scores with each resource. The groups swapped resources between the two tasks.
- Task 1. With Lenny: 71%. Usual resources: 54%. p = 0.065.
- Task 2. With Lenny: 63%. Usual resources: 68%. p = 0.634.
How students rated it
- Ease of use: Higher in both arms.
- Satisfaction: Higher in both arms.
- Quality of information: Higher in both arms.
- Ease of understanding: Higher in both arms.
- Engagement: Higher in both arms.
- Efficiency: Higher in one arm.
- Confidence applying information: Higher in one arm.
- Likely to use again: Higher in one arm.
- Depth of content: No difference.
- Critical thinking: No difference.
Significance at p below 0.05, Wilcoxon signed rank, within each arm.
What students said
if it was a single answer, then the chatbot was better
Chatbot is better for specific questions
more useful if references are included
Googling and using notes enhanced critical thinking instead of [using] the chatbot
best to train it to be tailored to [the] curriculum
generate Ankis, questions, and diagrams from PowerPoint
Focus group participants, quoted in the paper.
Discussion
Students described a clear role for an AI tutor: finding information quickly, clarifying a specific question and helping them check understanding. Our design priority is to pair that convenience with visible sources and opportunities to reason through a problem.
Context
Twenty first year medical students each completed an anatomy task with Lenny and another with conventional resources. Fifteen joined focus groups. The brief tasks assessed immediate performance; longer term learning remains a question for further study.
Ng, Siu, Han, Ho, Sun, Markiv, Knight and Gill Sagoo. Evaluating a Custom Chatbot in Undergraduate Medical Education: Randomised Crossover Mixed Methods Evaluation of Performance, Utility, and Perceptions. Behavioral Sciences 15(9):1284, 2025.
Read the paperConference papers
Journal of Anatomy
Anatomical Society Winter Meeting 2025, Journal of Anatomy, 2026
Perceived efficiency and test performance moved in opposite directions in an exploratory analysis of AI assisted anatomy learning.
- The analysis compared each student's change in perceived efficiency with their change in score when using Lenny and conventional resources.
- Higher confidence was associated with higher perceived information accuracy. Feeling that performance had improved also went with feeling more efficient.
- These associations help frame a question for education research: how closely does the experience of using AI reflect what a student has learned?
Changes that went together
Bars show associations between changes, with 95% intervals. Left of zero means the changes moved in opposite directions.
- Felt more efficient, scored lower. Estimate -0.39, 95% interval -0.92 to -0.09.
- Scored lower, felt more efficient. Estimate -0.68, 95% interval -1.59 to -0.11.
- Thought they did well, felt more efficient. Estimate 0.82, 95% interval 0.22 to 1.42.
Why we study this
Students' perceptions and task performance offer different views of learning. Studying both can help educators evaluate AI tools and help students assess their own understanding. This exploratory analysis used the original twenty student trial; it identifies associations and does not establish which change caused another.
Ng, Siu, Han, Ho, Sun and Gill Sagoo. When Easy Feels Effective: Perception Performance Mismatch in AI Assisted Anatomy Learning. Anatomical Society Winter Meeting, Anglia Ruskin University, December 2025. Journal of Anatomy, 2026.
Read the abstractICERI 2024
ICERI 2024, Seville, 2024
In preliminary feedback, students gave the evaluated AI written questions higher ratings for quality and for testing reasoning beyond recall.
- Students assessed the questions and the explanations for every option, including their clarity and relevance to the teaching material.
- Participants rated the improved question sets as better at testing reasoning beyond factual recall.
What this means for question writing
Students' feedback covered both what a question asks them to think through and how well its explanation teaches. The published abstract reports preliminary ratings from thirty preclinical medical students.
Ho, Markiv, Ng, Han, Gasa, Muthukumar, Lau, Sun and Gill Sagoo. Enhancing AI Generated Single Best Answer Questions in Medical Education: A Study on the Effectiveness of Prompt Chaining. ICERI 2024 Proceedings, p. 9761.
Read the abstractIFAA 2024
IFAA 2024, Anatomy and Cell Biology supplement, 2024
An early qVault study connected students' own teaching material to personalised questions and progress tracking.
- Students could upload course notes and practise questions based on that material, linking revision to the curriculum they were studying.
- The abstract reported preliminary improvements in production reliability and speed, with consistent question formatting as production expanded.
Why curriculum fit matters
Questions based on a student's teaching can make practice relevant to their course. Progress tracking helps students follow that practice across a term. This abstract describes early development of the platform; it does not report a comparison of student learning outcomes.
Ng, Markiv, Han, Ho, Gasa, Muthukumar, Lau and Gill Sagoo. Accelerating Medical Learning: AI Powered Personalised SBA Question Generation for Enhanced and Sustainable Progress Tracking. IFAA 2024. Anatomy and Cell Biology 57(3) supplement, abstract A0700.
Funded by the King's College Teaching Fund for generative AI.
Read the abstract bookTalks
qVault at the AI in Education Innovation Showcase
King's College London, 2026
Transforming Medical Education with qVault.ai
RIDE 2025, University of London, 2025
qVault.ai: Student Staff Co Design and Personalised Learning
12th Change Agents' Network Conference, University of Plymouth, 2025
Inclusive OSCE Reform: AI Case Generation and Real Time Communication Simulation
King's Academy Education Conference, 2025
Transforming Education through qVault.ai: Staff, Students and Beyond
AI in Education Showcase, King's College London, 2025
Awards and funding
Finalist, King's Engaged Research Awards
Integrated Education, Research and Service, 2026
Finalist, King's Engaged Research AwardsSPARK Innovation
King's Innovation Catalyst, 2026
College Teaching Fund: TASK
King's Academy, 2025–26
College Teaching Fund: Race Equity and Inclusive Education
King's Academy, 2024–25
Student Partnership Impact Award
SEDA and Jisc, 2024
Student Partnership Impact AwardCollege Teaching Fund: AI
King's Academy, 2023–24
In progress
Responsibility Inversion: Bottom Up Innovation in Generative AI Adoption, A Case Study with qVault.ai at King's College London
Book chapter, Springer Nature. Accepted for publication.
Explanations that teach
Feedback that turns mistakes into learning.
Knowing what you know
Metacognition, measured across every answer.
Second thoughts
The signal hidden in a change of mind.
Learning beyond the score
Measuring learning that lasts beyond the exam.
Every student's trajectory
How knowledge grows, one student at a time.
AI written assessment
Evaluating AI written questions with students.
Ethics and data
Research approvals
Existing King's studies: LRM 25/26 40754. New studies are assessed separately.
ICO registration
ZC218575
Reporting
Combined figures only, with small groups merged so no one stands out
Your choice
Told at sign up, opt out by email at any time
AI training
Student content is never used to train models