Nutrisha

Set up preferences
Diet Plan
Meal Upload
Meal Analysis
Dark mode

Nutrisha combines generative AI with MeldRx's FHIR platform to prevent adverse food-health interactions and provide personalized nutrition guidance, demonstrating significant potential for improving patient outcomes through smart dietary recommendations.

Inspiration

I am personally interested in nutrition and helping people get better at following advice from their doctors!

What it does

📅 Smart Diet Planning

Health-Aware Recommendations: Automatically considers:
- Active health conditions
- Current medications
- Recent lab results
- Dietary restrictions
- Food allergies

🔍 AI-Powered Food Analysis

Intelligent Image Recognition: Upload meal photos for:
- Precise nutritional breakdown
- Medication interaction warnings
- Health condition considerations
- Smart portion recommendations

🎨 User Experience

Intuitive interface with light/dark modes
Seamless health data integration
Real-time recommendations

Healthcare Impact

Prevention of adverse food-medication interactions
Improved medication timing compliance
Enhanced dietary adherence through personalization
Accessible nutrition guidance for patients with complex health conditions

Predictive Accuracy

Limited to the underlying accuracy of gpt-4o-mini
Limited to the underlying accuracy of the FHIR resources

How we built it

Starting from MeldRx's React template, I implemented the SMART on FHIR launch flow. Next, I wrote FHIR queries to get the necessary data and filtered it for relevant parts (like diabetes condition, weight, height Observations). Then, I started to experiment with generating nutrition plans from this data using OpenAI's API. Lastly, I explored using the Vision API to analyze images of meals.

System Diagram

Diagram

🔬 Advanced Integration Features

Comprehensive FHIR Integration
- Utilizes 6 critical FHIR resources for complete health context
- Real-time synchronization with health records
- Intelligent processing of clinical data for dietary impact
State-of-the-Art AI Implementation
- OpenAI gpt-4o-mini for intelligent diet planning
- OpenAI gpt-4o-mini for real-time food analysis
- Custom prompt engineering with structured JSON schema responses
- Advanced vision-language modeling for food recognition
Predictive Healthcare Features
- Anticipates medication-food interactions
- Predicts nutritional needs based on lab trends
- Suggests preventive dietary modifications

Challenges we ran into

Many! The MeldRx React template took some modification to get started. I made many API requests and had to ask for an increase. The OpenAI calls were tricky to figure out because I used structured outputs - and the responses were deeply nested.