VisionVine

AI-Powered Alcohol Label Verification

Author: Shadi Kabajah
Email: skabajah@icloud.com
Date: August 7, 2026
Live Demo: https://visionvine.onrender.com

Overview

VisionVine is a web application that uses AI (Groq’s Qwen 3.8-27b vision model) to extract and validate information from alcohol beverage labels. It is designed to assist TTB compliance agents in verifying that labels meet regulatory requirements, reducing manual review time from minutes to seconds.

Features

Upload 1–3 images per product (front label, back label, angle shots)
Supports:
- Single label image (one picture)
- Multiple images (2–3 pictures of the same product)
- Combined label (multiple labels in one image)
AI-powered text extraction (brand, class/type, ABV, net contents, government warning)
Automatic beverage type detection (spirits, wine, beer)
Validation of required fields
Government Warning header check (must be in ALL CAPS)
Pass/Fail results with field-by-field status
Clean, modern UI with Material Icons
Responsive design for desktop and mobile

Demo Example

The screenshots/ folder contains example results and test label images you can download and use to test the app.

File	Description
`result-1.png`	PASS result — single label image
`result-2.png`	PASS result — two separate images (front + back)
`result-3.png`	PASS result — one image containing two labels (combined)
`sample_1.jpg`	Sample label image 1 (distilled spirits)
`sample_2a.jpg`	Sample label image 2a (front label)
`sample_2b.jpg`	Sample label image 2b (back label)
`sample_3.png`	Combined label image (front + back in one)

Gallery Preview

Click any image to expand (opens in new tab).

Input	Output
_{Single label image}	_{✅ PASS — single label}
_{Two separate images (front + back)}	_{✅ PASS — two images}
_{One image containing two labels (combined)}	_{✅ PASS — combined label}

Tech Stack

Component	Technology	Homepage
Backend	Python + FastAPI	fastapi.tiangolo.com
AI / OCR	Groq API (Qwen 3.8-27b)	console.groq.com
Icons	Google Material Icons	fonts.google.com/icons
Frontend	HTML + CSS + JavaScript	—
Hosting	Render	render.com
Source Management	GitHub	github.com

UX Path

User lands on the page and sees a clean split-screen layout.
On the left panel, the user drags and drops 1–3 images of an alcohol label (or clicks the + button to browse).
The user clicks "Verify Label".
The app sends the image(s) to the backend, which calls Groq's vision model to extract:
- Brand name
- Class/Type
- ABV
- Net contents
- Government warning
- Beverage type
The extracted data is validated against TTB requirements.
Results appear on the right panel with:
- Field-by-field PASS/FAIL status
- Government Warning header check (must be in ALL CAPS)
- Overall PASS or FAIL decision
The user can click "Start Over" to reset and test another label.

Assumptions

Supports 1–3 images per product
Focus on distilled spirits labels (based on the sample provided in the assessment)
Government Warning check only verifies the header is in ALL CAPS (not the full text)
The app is a prototype and does not cover all legal variations
Images should be clear enough for OCR (JPG, PNG, WebP supported)

Limitations

Batch upload for multiple products not yet implemented
No matching against application form data
Only one image is processed per request (the first uploaded)
Wine and beer labels may have lower OCR accuracy due to decorative fonts

Free Tier Notes:

The app runs on Render's free tier — if inactive for 15 minutes, it goes to sleep. It will wake up automatically on the next visit (may take 5–15 seconds).
Groq API is used on the free tier — token limits apply (200,000 tokens per day). If exceeded, you may see a rate limit error.

File Structure

  visionvine/
  ├── main.py
  ├── requirements.txt
  ├── screenshots/
  │   ├── result-1.png
  │   ├── result-2.png
  │   ├── result-3.png
  │   ├── sample_1.jpg
  │   ├── sample_2a.jpg
  │   ├── sample_2b.jpg
  │   └── sample_3.png
  ├── static/
  │   ├── index.html
  │   ├── app.js
  │   ├── cleaner.js
  │   ├── style.css
  │   └── icon.svg
  └── README.md