Diagnosify - 1st place (€10,000 prize) Hackathon

AI-powered clinical simulator for medical education — built at a 24-hour hackathon, won 1st place (€10,000 prize) at Zoku Amsterdam in the Tech Europe.

Diagnosify lets medical students practice diagnostic reasoning through immersive, natural-voice conversations with AI-generated patients. Instead of reading textbooks, students interrogate a realistic patient, order physical examinations, and reach a diagnosis — all in a risk-free environment.

---

The Problem

Medical schools produce doctors who know anatomy and pharmacology but freeze in their first real clinical encounter. The gap between textbook knowledge and bedside manner is well-documented — and existing simulators are either expensive, text-only, or limited to scripted cases. Diagnosify bridges that gap with dynamic, voice-driven cases that respond naturally to whatever the student asks.

---

Features

• Voice-driven interaction — students speak naturally using high-accuracy medical STT (Deepgram Nova 3 Medical), optimized for clinical terminology
• Dynamic case generation — Google Gemini 2.5 Flash generates unique, medically accurate patient scenarios across Gastroenterology, Cardiology, Pulmonology, Neurology, and more
• Interactive physical examination — students must actively request vital signs, auscultation, palpation, and other exam maneuvers to uncover findings
• Realistic patient avatars — dynamically generated patient images (Fal.ai Flux) matched to the case demographics
• Immediate feedback — structured post-case review of diagnostic reasoning and missed findings

---

Tech Stack

Layer	Technology
Framework	Next.js 15 (App Router)
LLM	Google Gemini 2.5 Flash
Speech-to-Text	Deepgram Nova 3 Medical (via SLNG)
Image Generation	Fal.ai (Flux Schnell)
Styling	Tailwind CSS
Icons	Lucide React

---

Setup

Prerequisites

• Node.js v20+

Environment Variables

Create .env.local:

GEMINI_API_KEY=your_gemini_key
FAL_API_KEY=your_fal_key
NEXT_PUBLIC_SLNG_API_KEY=your_slng_key

Install & Run

npm install
npm run dev

This starts both the Next.js server (port 3000) and the SLNG WebSocket proxy (port 3001).

---

How It Works

1. The student selects a specialty and difficulty level
2. Gemini generates a complete patient case: demographics, presenting complaint, history, exam findings, and final diagnosis — all internally consistent
3. The student interviews the patient using voice; Deepgram transcribes in real-time with medical vocabulary support
4. Physical exam actions (check vitals, auscultate heart, palpate abdomen) reveal findings progressively — students can't skip straight to the answer
5. When the student is ready, they submit a diagnosis with reasoning

---

Built in 24 hours at Hackathon Zoku, Amsterdam — April 2026.