Amelie

• 

  Daily check in

• 

  Dashboard

• 

  AI Twin

Inspiration

Rare disease patients live in constant uncertainty. Symptoms fluctuate daily, appointments are brief and infrequent, and most patients struggle to accurately communicate weeks of complex health changes in a 15-minute consultation. Many also experience deep isolation because only a small number of people share their condition. We were inspired by a simple but powerful question: What if patients had an intelligent system that tracked, understood, and translated their daily experience into meaningful insights? Amelie was built to give rare disease patients clarity, structure, and confidence in managing their health.

What it does

Amelie is a full-stack web application designed specifically for rare disease patients. It provides: (a) Daily health check-ins (pain, energy, sleep, mood, mobility, red flags) (b) AI-generated personalized daily plans (c) A 7-day predictive risk forecast (d) Rosetta — an AI health companion (e) Smart safety alerts (f) Clinician-ready PDF health reports (g) Anonymous condition-specific community groups Unlike traditional health trackers that only show historical data, Amelie predicts risk trends and translates patient data into actionable insights. It helps patients prepare for appointments, recognize worsening patterns early, and connect with others who truly understand their condition.

How we built it

Frontend:

1. React
2. Vite
3. Tailwind CSS
4. Backend
5. Node.js + Express
6. Socket.io for real-time community chat
7. JWT authentication with bcrypt

Database:

1. sql.js (WebAssembly SQLite) for zero native setup

AI Features:

1. OpenAI GPT-4o via streaming Server-Sent Events
2. Personalized daily plan generation
3. AI twin (Rosetta) with contextual system prompts
4. GPT-generated clinical report PDFs

Custom Backend Logic:

1. Rule-based alert engine
2. Weighted mathematical 7-day risk forecast
3. Linear regression trend calculations
4. Risk dampening over time

The forecast and alert systems run purely on backend logic — no AI calls — ensuring speed and reliability.

Challenges we ran into

1. Designing a risk prediction formula that felt medically conservative yet meaningful
2. Ensuring AI responses were contextual, not generic
3. Preventing over-medicalization or unsafe AI suggestions
4. Balancing technical complexity with a clean, simple UI
5. Making forecasts feel helpful — not anxiety-inducing

We had to carefully separate:

1. AI-driven guidance
2. Deterministic safety rules
3. Predictive mathematical modeling

This architecture was critical for both safety and performance.

Accomplishments that we're proud of

1. Building a real-time AI twin that understands personal health baselines
2. Creating a backend-driven predictive forecast under 5ms response time
3. Designing a rule-based alert engine independent of AI
4. Generating clinician-ready PDF reports automatically
5. Implementing anonymous peer communities with live messaging

Most importantly: We built something that feels empathetic — not just technical.

What we learned

1. In health tech, safety > cleverness
2. AI must be guided by strict boundaries
3. Predictive modeling requires thoughtful dampening to avoid false certainty
4. Simplicity in UI builds trust
5. Real impact comes from reducing cognitive load for vulnerable users

We also learned how powerful structured data becomes when paired with intelligent interpretation.

What's next for Amelie

1. Integrating wearable data (Apple Health, Garmin, Fitbit)
2. Adding longitudinal analytics beyond 7 days
3. Clinician dashboards for shared access
4. HIPAA-compliant cloud deployment
5. More rare-disease-specific forecasting models
6. Expanding multilingual support

Long term, we envision Amelie becoming a structured intelligence layer between patients and providers—improving communication, early intervention, and quality of life.

Built With

• css
• express.js
• node.js
• openaiapi
• react
• tailwind
• vite