2026🥇 1st placeHackathon
MedScribe
AI medical scribe — 1st place, Decoding Data Science 2026 Hackathon
Walk-out-with-the-note medical scribe. Clinician dictates after a visit, Whisper transcribes, Claude drafts both the SOAP note and a plain-English patient handout. Built in one hour at the Decoding Data Science 2026 Hackathon — 1st place.
Problem
Clinicians lose the thread between the visit and the chart. By the time they sit down to write the SOAP note, the sharpest details are already fading — and the patient leaves without a clear handout of what to do next. Most scribe tools target the visit itself; nobody catches the 60 seconds after.
Role
1-hour hackathon build with the team. Wired browser audio → Whisper for transcription, Claude Sonnet 4.6 via the Vercel AI Gateway for structured drafting of two documents in parallel — SOAP note for the chart, plain-English handout for the patient — and react-pdf for clinician-signed exports. Post-hackathon I refined the landing page, the explainer video, and the privacy story.
Outcome
1st place at Decoding Data Science 2026 from a 60-minute build. Live at trymedscribe.vercel.app. The shape of the product clarified mid-sprint: a scribe that produces two artefacts from one dictation — chart-grade SOAP for the clinician and plain-English handout for the patient. The constraint forced ruthlessness about what mattered: record, transcribe, draft, sign, export. Lesson: under a 60-minute clock, the right move is to ship the smallest end-to-end loop and trust the polish pass to come later.
Stack
- Next.js 16
- React 19
- Whisper
- Claude Sonnet 4.6
- Vercel AI Gateway
- react-pdf
- Tailwind
- Remotion
- Vercel
Next project
Sara →
Voice agent for a fictional fractional real estate platform