NutriSight

NutriSight helps people choose better food before they eat, using AI camera vision, personal goals, meal memory, and future AI glasses.

Comment

Inspiration

NutriSight was inspired by a simple everyday moment: standing in front of food and not really knowing what the best choice is right now.

Most people make food decisions in the moment — at a café, restaurant, supermarket, or at home — but only understand the nutritional impact afterwards, if they track it at all. Traditional nutrition apps usually answer the question: “What did I eat?” NutriSight is designed to answer a more useful question before the decision is made: “What should I choose right now?”

The idea was also inspired by conversations around visual nutrition education, including the concept of making hidden sugar, carbohydrate, and glucose impact easier for everyday people to understand. NutriSight builds on that idea, but expands it beyond sugar into a broader real-time nutrition assistant for calories, carbohydrates, protein, fiber, salt, fat, personal goals, and daily balance.

The core vision is simple:

NutriSight helps people understand food before they eat it, not after.

What it does

NutriSight is a real-time nutrition assistant for phones today and future AI glasses tomorrow.

Using AI camera vision, personal goals, meal memory, and gamified motivation, NutriSight helps people make healthier and more informed food choices in the moment.

A user can scan or compare:

  • A meal
  • A menu
  • A café counter
  • A supermarket product
  • A restaurant choice

NutriSight then estimates key nutrition information such as:

  • Calories
  • Sugar
  • Carbohydrates
  • Protein
  • Fiber
  • Salt
  • Fat
  • Overall nutrition impact

Users can choose what they want to optimize for, such as:

  • Lose weight
  • Reduce sugar
  • Stabilize blood sugar
  • Eat more protein
  • Eat more fiber
  • Reduce salt
  • Reduce calories
  • Make healthier choices overall

Instead of giving generic advice, NutriSight uses meal memory to understand what the user has already eaten during the day. This makes the recommendation personal.

For example, if a user is choosing between a salad and a donut, NutriSight can explain which option fits better with their current daily balance and why.

NutriSight is not designed to shame users or make food boring. The goal is balance, not restriction. Sometimes the best choice is a salad. Sometimes there is room for a donut, beer, or treat if it fits the user’s goals and daily balance.

Gamified motivation

NutriSight is designed to be engaging, not restrictive.

To help users stay motivated, the product includes gamified features such as:

  • Daily streaks
  • Health scores
  • Progress badges
  • Small daily challenges
  • Positive feedback when users make better choices
  • Simple visual explanations like sugar-cube or carb-impact estimates

The goal is to make healthy eating feel understandable, rewarding, and motivating — not boring or judgmental.

How NutriSight uses AI

NutriSight is designed around an AI-agent workflow rather than a static nutrition calculator.

The core AI components are:

  1. Food Recognition Agent
    Identifies food items from images, menus, or live camera views.

  2. Menu Understanding Agent
    Reads menu items and compares food options.

  3. Nutrition Estimation Agent
    Estimates calories, sugar, carbs, protein, fiber, salt, and fat.

  4. Personal Meal Memory Agent
    Tracks what the user has already eaten during the day.

  5. Recommendation Agent
    Suggests the better choice based on the user’s personal goal and current nutrition balance.

  6. Gamification Agent
    Creates streaks, health scores, progress feedback, and small challenges.

  7. Safety & Disclaimer Agent
    Ensures the app gives estimated guidance, not medical advice.

A simplified version of the recommendation logic is:

$$ Recommendation = f(FoodData, UserGoal, MealHistory, DailyBalance) $$

Where:

  • (FoodData) = estimated nutrition from the scanned food
  • (UserGoal) = the user’s chosen focus
  • (MealHistory) = what the user has already eaten
  • (DailyBalance) = remaining calories, sugar, protein, fiber, salt, and other targets

How we built it

NutriSight was built as a polished MVP demo for an AI-first product concept.

The prototype demonstrates the core user journey:

  1. Choose a personal nutrition goal
  2. Scan or compare food options
  3. Review estimated nutrition impact
  4. See what has already been eaten today
  5. Receive a personalized recommendation
  6. Stay motivated through gamified progress

The project is designed to use:

  • Gemini API for multimodal food and menu understanding
  • Google AI Studio for prompt testing and AI behavior design
  • Google Cloud as the planned scalable infrastructure
  • Firebase / Firestore for user accounts, meal memory, and daily intake history
  • Cloud Storage for food images and menu uploads
  • React and TypeScript / JavaScript for the MVP interface
  • Nutrition databases / APIs for nutrition estimation
  • AI agents for food recognition, menu understanding, nutrition estimation, meal memory, recommendations, gamification, and safety

Gemini is central to the product vision because NutriSight needs to understand more than a single food item. It needs multimodal reasoning: food images, menu text, user goals, meal history, context, and natural-language explanations.

The current MVP demonstrates the intended Gemini-powered experience. The next production milestone is to connect the live demo more deeply to Gemini API calls, trusted nutrition databases, Firebase meal memory, and real user inputs.

Challenges we faced

One of the biggest challenges was balancing simplicity with accuracy.

Nutrition estimation from images is never perfect. Portion sizes, hidden ingredients, cooking methods, sauces, oils, and restaurant variations can all affect the result. Because of that, NutriSight presents outputs as estimates, not medical facts.

Another challenge was avoiding overly narrow positioning. The first idea focused heavily on sugar and sugar-cube visuals, but the project evolved into a broader nutrition assistant. Users should be able to choose what matters to them — sugar, calories, protein, fiber, salt, carbs, or overall healthier choices.

We also had to think carefully about health responsibility. NutriSight is not designed to diagnose, treat, or replace doctors, dietitians, or medical devices. It provides estimated guidance to help users make more informed everyday food decisions.

A third challenge was making the product feel human. A nutrition assistant should not act like a strict diet police. NutriSight should help users understand trade-offs and make better choices, not perfect choices.

What we learned

We learned that the strongest version of this idea is not just a calorie tracker.

The real value is helping people at the decision point.

Most food apps answer:

“What did I eat?”

NutriSight answers:

“What should I choose right now?”

We also learned that personal memory is essential. A donut is not always “bad,” and a salad is not always “perfect.” The best recommendation depends on the user’s goal, what they already ate, and what they need for the rest of the day.

We also learned that AI can make nutrition more understandable when it translates complex data into simple, visual, personalized guidance.

What’s next

The next steps for NutriSight are:

  • Add a clearer live Gemini-powered recommendation flow
  • Improve real food and menu recognition
  • Connect to trusted nutrition databases
  • Add user accounts and personal meal history
  • Build a stronger daily goal and nutrition balance system
  • Add beta user testing
  • Collect feedback and public testimonials
  • Develop premium reports and coaching features
  • Explore B2B pilots with health coaches, clinics, fitness centers, and insurance partners
  • Prepare for future integration with audio-first AI glasses

Vision

NutriSight turns complex nutrition data into clear, practical advice.

The long-term vision is to create a real-time food intelligence layer for everyday life — on phones today and AI glasses tomorrow.

NutriSight helps people make better choices, not perfect choices — before they eat, not after.

Built With

Share this project:

Updates

posted an update —

NutriSight MVP is live

I have launched the first MVP demo of NutriSight:

https://nutrisight.lovable.app

NutriSight is a real-time nutrition assistant that helps people understand food before they eat it. The goal is to move nutrition guidance from “after the meal” to the moment where the decision is actually made.

The current demo shows the core product direction:

  • AI camera vision for food and menu understanding
  • Personal goal selection, such as reducing sugar, losing weight, eating more protein, or making healthier choices
  • Meal memory, so recommendations can consider what the user has already eaten during the day
  • Simple nutrition insights, including calories, sugar, carbs, protein, fiber, salt, and overall nutrition impact
  • Gamified motivation through health scores, streaks, challenges, and progress feedback
  • A future vision for AI glasses, where users could receive private nutrition guidance while looking at a menu, café counter, supermarket shelf, or meal

One of the key insights behind NutriSight is that people do not only need food tracking. They need decision support before they eat.

Next steps:

  • Add a clearer Gemini-powered recommendation flow
  • Improve food/menu analysis
  • Connect stronger meal-memory logic
  • Collect beta feedback from real users
  • Continue refining the product for the Build with Gemini XPRIZE Hackathon

NutriSight helps people understand food before they eat it, not after.

Log in or sign up for Devpost to join the conversation.