Functionality and Architecture Description

1. Access and Monetization (The Funnel)

• Guest Mode (Free Entry):
  • The user can start using the app without registration.
  • Limit: 10 free scans per device (device ID).
  • Goal: Show the value of "Shazam for products" instantly.
• Registration & Subscription:
  • After 10 scans — mandatory registration (Google/Apple/Email).
  • Offer: 1 month free upon subscription.
• Pricing Plans (Annual Payment):
  • Starter Plan: Free (limit: up to 10 scans per month).
  • Basic Plan: $49.99 / year (limit: up to 1000 scans per month).
  • Premium Plan: $99.99 / year (limit: up to 5000 scans per month).

2. Onboarding and Personalization

• Setting Limits (User Tags):
• The user selects (simply taps on the tags) critical parameters for them (in English for AI):
  • Food: Gluten-free, Lactose-free, Nuts, Eggs, Seafood, Vegan, Sugar-free.
  • Care & Home: Fragrance-free, Sulfate/Paraben-free, Sensitive Skin, Cruelty-free.
• Location: Request access to GPS for Grounding (search for prices and reviews in a specific country/city).

3. Core Loop (Scanning Cycle)

• Shooting Process:
  1. The user takes a photo of 1 product. (or adds from the gallery)
  2. Buttons: "Analyze" (if there is one product) or "+ Add product to compare" (if they want to compare).
  3. When comparing, the user takes separate photos of the 2nd and 3rd products. Analysis occurs asynchronously for speed.
• AI Backend (The Engine):
  • Model: Gemini 2.5 Flash (Temp = 0.0).
  • Grounding: Mandatory Google Search Retrieval considering the user's location.
  • Mode Detection: AI receives an array of photos. If there is one photo — Product Passport mode. If 2-3 — Comparison Matrix mode.

4. Analysis Results (UI/UX)

• Photos of products with prices in local networks are displayed.
• Traffic Light System: Visual color codes (Green/Yellow/Red).
• Why NOT to Buy: Key block. AI focuses on negative aspects: inflated price, harmful additives, poor consumer rating.
• Truth Protocol: Display of the number of analyzed sources (review_count) and direct quotes-excerpts from Reddit, Amazon, and local marketplaces.
• Verdict (Main Part): Final recommendation (in comparison mode, AI clearly indicates the winner or its absence and why).

5. Additional Functionality (Utility)

• Scan History: Storing the history of all scans (available only to registered users).
• Favorites: List of "favorite" products for quick access.
• Manual Search: Text search (if the camera did not recognize the brand or the label is damaged).
• Feedback: Ability to report incorrect data for system retraining.

6. Technical Stack (Backend Requirements)

• Framework: FastAPI (Python), asynchronous work with Gemini API.
• Database: Supabase or PostgreSQL (storing profiles, scan limits, and history).
• AI Integration:
  • Input: image_bytes, location, user_tags.
  • Output: Structured JSON for instant table and card building on the frontend.