Scan The Lie

Food labels lie. Gemini 3 catches them. Your right to know what's in your food shouldn't be hidden under false labels and marketing lies

Comment

Inspiration

In a world where "all-natural" claims hide artificial ingredients and "healthy" labels mask harmful additives, I was inspired by the daily struggle of families and individuals trying to make informed food choices. The packet food industry's smoke and mirrors approach to labeling where "natural flavors" often means anything but natural, and "organic" claims hide synthetic ingredients became my call to action.

My mission is to use technology as a tool for social justice, to empower consumers against deceptive marketing practices and to create a world where food labels tell the truth, not just a marketing story. Scan The Lie is my response to the growing need for transparency in the food industry, where the gap between marketing claims and actual ingredients has become a public health concern.

What It Does

Scan The Lie is an Gemini powered mobile application that:

  • Dual Image Scanning: Captures both front (marketing claims) and back (ingredients) labels of packet food products for comprehensive analysis
  • Claim Verification: Uses Gemini 3 to verify each marketing claim against actual ingredients of the food product (Verified/Misleading/False)
  • Deep Ingredient Analysis: Provides detailed breakdowns including purpose, origin, risk level, controversies, and banned countries, of the ingredients listed in the product label.
  • Personalized Health Insights: Tailors analysis based on your allergies, dietary preferences, and health concerns. A Questionnaire will be present at the start to understand your health profile.
  • AI Health Chatbot: Context-aware assistant that answers follow-up questions about scanned products
  • PDF Report Generation: Creates professional, shareable analysis reports of each product you scan

How I Built It

  • Frontend: Flutter with Dart for cross-platform mobile development
  • AI Integration: Google Gemini 3 Flash Preview (multimodal vision + chat)
  • Local Storage: Hive NoSQL database for scan history and user preferences

Gemini 3 Integration

This application harnesses three powerful Gemini 3 features. The vision model analyzes dual product images in parallel, one capturing marketing claims, the other revealing actual ingredients performing text extraction and visual interpretation across varying label formats. Beyond simple OCR, Gemini 3 applies logical reasoning to verify truthfulness: for example, when a product advertises "No Artificial Colors" yet contains synthetic dye E102, the model identifies the discrepancy through knowledge based deduction in a single API call. Additionally, the system supports personalized analysis by injecting user health profiles directly into prompts .As a result, dietary preferences, allergies, and health goals become part of the reasoning context, allowing Gemini 3 to surface relevant warnings and recommendations unique to each individual user.

More Technical Documentation is provided in the zip folder attached to the judges.

Challenges I Ran Into

  1. Multimodal Precision: Ensuring Gemini 3 accurately extracts information from varied product label designs, fonts, and lighting conditions required extensive prompt engineering.

  2. Non Food Detection: Implemented validation to reject scans that aren't food products, preventing irrelevant analysis.

  3. Context Persistence: Maintaining product context in the chatbot while keeping responses concise required careful prompt design.

Accomplishments I'm Proud Of

What makes me most proud is the actual purpose behind my project. I built something that genuinely matters for public health. Every day, millions of people unknowingly consume products with misleading labels, and I've created a tool that puts the power of truth back in their hands. The fact that I could harness the incredible capabilities of Gemini AI, its multimodal vision, its reasoning, its ability to understand context and channel all of that into something that protects families, helps people with allergies stay safe, and exposes marketing deception... that's what keeps me going.

I'm also incredibly proud of fine-tuning Gemini to serve this specific purpose. Through careful prompt engineering, I shaped the AI to think like a consumer health advocate to be skeptical of marketing claims, to flag controversial ingredients, to understand dietary restrictions, and to communicate findings in a way that empowers everyday people.

What I Learned

  • Prompt Engineering at Scale: Crafting prompts that return consistent, structured JSON across thousands of different product types
  • Mobile First AI Integration: Managing API latency expectations and providing engaging loading experiences

What's Next for Scan The Lie

1) Community Features Planning to add a forum where users can discuss scanned products, share recommendations, and build a community of informed consumers.

2) Barcode Integration Adding UPC/EAN barcode scanning for instant product lookup and historical data comparison.

Built With

Share this project:

Updates