NutriScan

• 

  link to app!

• 

  Home page

• 

  Analysis page

• 

  Result page

• 

  Alternative Receipe

• 

  result

• 

  Goals page

• 

  home page 2

NutriScan is not just a calorie counter — it’s a decision-making assistant for food.

Using MeDo, I was able to:

~ design a realistic nutrition intelligence system ~ implement structured, scalable outputs ~ and build a polished, user-focused application

This project showcases how AI can be used not just for automation, but for meaningful, personalized health insights.

About

I built NutriScan, an advanced nutrition analysis web application that helps users understand what they’re actually eating — beyond just calories.

The core problem I wanted to solve is that most people either:

• don’t track nutrition at all, or
• rely on apps that provide generic or incomplete insights

NutriScan bridges that gap by allowing users to input either a dish name or a food image (processed into description) and receive a detailed nutritional analysis, including:

• Calories, protein, carbohydrates, and fats
• Hidden health risks such as added sugars, trans fats, high sodium, and ultra-processed ingredients
• Personalized recommendations based on user goals (weight loss, muscle gain, maintenance, diabetic- friendly, heart-healthy)

Additionally, the app goes a step further by suggesting healthier recipe alternatives. For example:

• High sodium dish → suggests low-sodium preparation methods
• High sugar dessert → recommends natural sweetener swaps
• Processed meals → suggests cleaner, whole-food versions

This transforms the app from just an analyzer into a practical decision-making assistant.

How I Used MeDo

I used MeDo as a co-builder and system designer, not just for generating code. I structured interactions around clear modules:

• Nutrition analysis logic
• Structured JSON outputs
• UI/UX system design

• Backend API flow (FastAPI + frontend integration) Instead of one-shot prompts, I iteratively refined:

• Prompt engineering for realistic nutrition estimation

• Detection logic for hidden unhealthy components

• Context-aware recommendation generation I also used MeDo to simulate real-world uncertainty, ensuring outputs remain practical and not overly optimistic.

Best Part MeDo Generated

The most impressive component is the intelligent nutrition + recommendation engine. It:

• Converts vague dish inputs into realistic nutritional estimates
• Detects non-obvious risks (like hidden sodium or processed ingredients)
• Provides specific, contextual insights instead of generic advice
• Suggests healthier alternatives and recipe modifications tailored to the dish

For example:

“High sodium due to processed sauce — consider a homemade version using herbs and lemon instead of salt.”

This level of actionable feedback makes the system genuinely useful.

Advanced Features & Extensions

📷 Image-based input (via preprocessed vision descriptions) 🧠 Structured JSON outputs for scalability 🎯 Goal-based recommendation engine 🍽️ Healthier recipe alternatives & ingredient swaps 💾 User system with saved history tracking

                                                 "Your plate has trust issues now" ......
                                                            THANK YOU ~_~

Built With

• llm
• medo
• supabase