Medical education tooling · personal project
Dr. Hank: a cited-source study assistant for medical students
Students asking a language model a Step 1 question get a fluent answer with no way to check it, and no path back to the resource they already pay for. The design goal here was the opposite: answer in the structure students actually study in, and end every answer at a citation they can open.
Status. Working prototype. Related work: I sit on the ScholarRx Student Advisory Committee and assist its AI development team.
This hub is still being built. The analysis is underway; the derivation, tables and figures will be posted here as they are finished, the way the completed projects on this site are. What is below is the design and the current state, written up as it stands.
The question
Can a language model be made genuinely useful for board study without becoming either a plagiarism engine or a confident source of unverifiable claims?
How it is being done
- Answers follow a fixed structure — a direct high-yield response, then, where relevant, a disease breakdown covering pathophysiology and clinical features.
- Every answer cites the relevant section of First Aid, AMBOSS or Rx Bricks and links out to it; content is pointed at rather than reproduced, which keeps the tool on the right side of both copyright and comprehension.
- Response length, format and detail are refined by A/B testing, with user preference on formatting sampled every ten to twenty interactions.
What comes next
The intended direction is a tool that identifies where a student is weak and generates against it — practice questions, spaced-repetition decks, case presentations and study plans — rather than only answering what is asked.
This page grows as the work does.