PureBloom

AI nutrition scoring, personalized to your health goals.

Comment

Inspiration

Every nutrition app tells you the same thing: calories, macros, generic ratings. But nutrition isn't one-size-fits-all. A high-fat avocado is excellent for someone seeking sustained energy, but less ideal for weight loss. We wanted to build something smarter—an AI nutritionist that understands your goals and scores foods accordingly.

When Gemini 3 launched with its advanced multimodal capabilities, we saw the opportunity to create truly personalized nutrition guidance. Not just calorie counting, but intelligent analysis that adapts to whether you're building muscle, clearing your skin, or optimizing gut health.

What it does

PureBloom gives every food a Bloom Score (0-10) personalized to your health goal. Users choose from six objectives: Weight Loss, Muscle Building, Clear Skin, Gut Health, Sustained Energy, or Heart Health.

Scan anything: Point your camera at packaged foods (barcode) or fresh meals (photo). Gemini 3 identifies ingredients and calculates your personalized score instantly.

AI Nutritionist Chat: Ask questions like "Is this good for my skin?" or "What should I eat before my workout?" and get goal-aware answers.

Track Progress: Monitor daily and weekly Bloom Scores to see patterns and stay motivated.

How we built it

Mobile App:

  • React Native + Expo for iOS development
  • Expo Camera for barcode and photo capture
  • Custom animations and haptic feedback

AI Backend:

  • Gemini 3 API for multimodal food recognition from photos
  • Gemini 3 for natural language AI Nutritionist conversations
  • Custom prompting to incorporate user's health goal context into every analysis

Infrastructure:

  • Supabase for authentication and database
  • RevenueCat for subscription management
  • Next.js website with waitlist

Bloom Score Calculation:

Gemini 3 weighs nutritional factors differently based on the selected goal:

Weight Loss     → low calorie density, high protein, high fiber
Muscle Building → high protein, moderate carbs, leucine content
Clear Skin      → low glycemic, antioxidants, omega-3s
Gut Health      → fiber, prebiotics, fermented foods
Energy          → complex carbs, B vitamins, iron
Heart Health    → low sodium, healthy fats, potassium

Challenges we ran into

Accurate Food Recognition: Getting Gemini 3 to reliably identify foods from user photos—especially complex dishes—required extensive prompt engineering. We iterated on system prompts to improve accuracy for mixed meals and regional cuisines.

Personalization at Scale: Calculating truly personalized scores meant building a framework that weights 15+ nutritional factors differently for each health goal. Balancing accuracy with response speed was challenging.

Making AI Responses Actionable: Early AI Nutritionist responses were too generic. We refined prompts to ensure responses reference the user's specific goal and provide concrete advice rather than general nutrition facts.

iOS App Store Compliance: Navigating Apple's guidelines for health apps and subscription UX required multiple iterations.

Accomplishments that we're proud of

  • Built a fully functional iOS app from concept to App Store submission in under 4 weeks
  • Achieved accurate food recognition across diverse cuisines and meal types using Gemini 3's multimodal capabilities
  • Created a scoring algorithm that makes the same food score differently based on personal health goals—true personalization
  • Designed an AI Nutritionist that maintains context about user goals and provides actionable, goal-specific advice
  • Implemented a clean, intuitive UI that makes nutrition tracking feel effortless rather than tedious

What we learned

Multimodal AI changes everything: Gemini 3's ability to understand images and text together enabled features that weren't possible before—identifying unlabeled foods, estimating portions, understanding meal context.

Personalization > Generic Advice: Users engage more when recommendations feel tailored to them. The same "8/10" score means more when it's calculated for your specific goal.

Prompt Engineering is Product Design: The quality of AI responses depends heavily on how you structure prompts. We spent as much time on prompts as on UI code.

What's next for PureBloom

  • Android launch
  • Meal planning with Bloom Score optimization
  • Integration with fitness trackers
  • Restaurant menu scanning with real-time scoring
  • Social features for sharing healthy finds

Built With

Share this project:

Updates