Smart FRCS AI: Teaching an Examiner to Listen

The FRCS Section 2 oral examination is spoken. A surgeon can know the clinical answer and still lose marks because it comes out unstructured, too slowly, or falls apart under the second question. Most preparation tools are written: they test recognition, not delivery. Rehearsing aloud usually means finding a consultant or a colleague with time to play examiner — and that is the scarcest resource a trainee has. The brief was to give candidates an examiner they can sit in front of at any hour: one that asks a clinical question, listens to a spoken answer, pushes on what was actually said, and then explains the result against the standard the real examination uses.

• The examination is spoken, but almost all preparation material is written.
• Practising with a real examiner depends on senior colleagues having time.
• Feedback has to map to the band scale candidates are actually marked on.
• The same rehearsal had to work on a laptop at home and a phone on a break.

As exclusive technical partner to algarra7 Ltd, the UK company that owns the product, I designed and built Smart FRCS AI as one product on three platforms. Each clinical case is a station. Before entering, the candidate chooses an examiner — Alden or Mira, two original examiner identities created for the product, holding one examination standard — and an answer mode: push-to-talk or continuous. Inside the station the examiner asks, listens while the answer is transcribed live, and follows up on the candidate's own words rather than reading from a fixed script. The station ends with a written report scored on the Intercollegiate 4–8 band scale and an annotated transcript the candidate can replay. A separate Keyword Practice mode, scored on coverage, lets candidates revise without spending viva minutes. The web app and the iOS and Android apps share one codebase and one account, with purchases through Apple, Google and the web.

Is the examiner reading from a script?

No. Each station starts from a clinical brief, but the follow-up questions are generated from what the candidate actually said in their previous answer.

Who are Alden and Mira?

Two original examiner identities created for Smart FRCS AI. The candidate chooses one per station; both hold the same examination standard.

Can you build a voice-based AI product like this for another field?

Yes. The hard parts — real-time voice, examiner logic grounded in a scope, structured scoring and multi-store billing — transfer to interviews, language assessment and professional training. Tell me what has to be assessed and I will tell you honestly how far AI should go.