Zphones screenshot
Zphones screenshot
Zphones screenshot

Zphones

Smarter phone inventory through computer vision.
View Img
Project Description
Summary

ZPhones is an AI-powered inventory management and phone recognition platform designed to automatically identify smartphones from images, estimate their market price, and manage stock in real time.

The system combines deep learning computer vision models with a modern web interface to streamline product registration and inventory operations for mobile device businesses.

ZPhones follows a modular architecture with a FastAPI backend, React frontend, and MongoDB database, ensuring scalability, performance, and clear separation of concerns.

Core Features & User Experience
  • Drag & drop phone image upload with instant preview
  • Automatic phone detection with confidence-based validation
  • Two AI classification pipelines: Brand → Model pipeline or Direct Model classifier
  • Manual classification override for low-confidence or edge cases
  • Storage selection, condition (new/used), and AI-based price prediction
  • Real-time inventory listing with images, prices, quantities, and dates
  • Fully responsive UI optimized for desktop and mobile usage
AI Detection & Pricing Engine

ZPhones integrates multiple deep learning models to analyze phone images and predict accurate pricing.

The image analysis pipeline includes:

  • A splitter model to detect no-phone, multiple-phones, or single-phone cases
  • A brand classifier followed by brand-specific model classifiers
  • An alternative direct classifier for single-step model prediction

For pricing, a regression model estimates phone value based on:

  • Brand and model
  • Storage capacity
  • Condition (new or used)
ResultConfidence thresholds are applied to ensure reliable predictions and prevent invalid results.
Backend & Data Management
  • FastAPI backend with async MongoDB integration
  • GridFS used for efficient image storage and streaming
  • RESTful API design covering analysis, pricing, inventory, and image delivery
  • Structured request and response models using Pydantic
  • Fully asynchronous database and file operations for high performance
Security & Reliability
  • Environment-based configuration for sensitive credentials
  • CORS configuration for controlled frontend access
  • Validation of inputs, confidence thresholds, and inventory updates
ResultThe system ensures data integrity while maintaining a smooth and secure user workflow.
Technical Stack
  • Frontend: React (Hooks, Fetch API, responsive UI)
  • Backend: FastAPI, Motor (async MongoDB), GridFS
  • AI & ML: TensorFlow / Keras, Scikit-learn (regression), Joblib
  • Database: MongoDB with GridFS image storage
ResultZPhones demonstrates a complete end-to-end AI-driven product recognition system, bridging machine learning, backend engineering, and modern frontend design into a single cohesive solution.
Added Value

This project provided hands-on experience with integrating pretrained deep learning models and computer vision pipelines into a real production-style application. It strengthened teamwork and communication skills by coordinating tasks, discussing design decisions, and aligning backend, frontend, and AI components among three team members. The workload was clearly split between AI model training and inference, backend API and database management, and frontend user interface development, reflecting real-world team structures. ZPhones represents a complete end-to-end system including AI, API, database, and frontend, mirroring modern software development practices commonly used in today’s tech companies.

Tags

Artificial Intelligence

Integrates pretrained AI models for automated phone recognition, confidence scoring, and intelligent decision making.

Computer Vision

Uses deep learning and image analysis to detect and classify smartphone models from uploaded images.

Team Collaboration

Developed by a three-member team with clear role distribution across AI, backend, and frontend development.

Languages & Competences

Python

Core language used for AI model loading, inference pipelines, data processing, and backend logic.

  • Implemented computer vision preprocessing pipelines for image-based classification
  • Integrated pretrained deep learning models for inference
  • Managed AI-related data flows and confidence thresholds

FastAPI

High-performance backend framework used to expose AI services and inventory APIs.

  • Designed REST APIs for image analysis, classification, and inventory management
  • Implemented multipart file upload handling for image inference
  • Structured API responses with validation and error handling

TensorFlow / Keras

Deep learning framework used for training, fine-tuning, and deploying phone recognition models.

  • Loaded and served pretrained CNN models for production inference
  • Applied confidence-based filtering and multi-stage classification pipelines

React (JavaScript)

Frontend framework used to build an interactive UI for AI detection and inventory control.

  • Built image upload and preview components with drag-and-drop support
  • Integrated frontend with backend APIs using fetch
  • Managed application state for detection results and inventory actions