Teezily, the French-based but globally operating eCommerce company with a unique business model, was looking for an AI solution to improve its performance by automating processes.
Addepto Team delivered the computer vision engine for image detection, and proved itself to be a reliable partner in implementing customized AI models into existing eCommerce infrastructure.
Case Study Shortcut
Challenge
Uncontrolled Image Quality in User Uploads
Teezily’s open platform empowered users to upload their own designs, but this led to frequent issues with low-quality images, including blurriness, poor cropping, and distortion. These flaws directly affected the print quality of final products, resulting in customer dissatisfaction and production inefficiencies.
Lack of Scalable Manual Quality Control
Scaling manual reviews of every uploaded image was not viable due to high costs and time constraints. Moreover, such an approach conflicted with Teezily’s mission of offering a seamless and frictionless creative experience for users.
No Internal AI Expertise to Build a Solution
Without an in-house AI team, Teezily lacked the technical resources to develop an automated quality control tool. While the team recognized the potential of AI, they needed external consulting to navigate the complexities of computer vision model development and deployment.
Goal
Teezily was focused on enhancing the caliber of products available to customers.
- Automate Image Quality Assessment at Scale
- Maintain Product Quality Without Hindering User Creativity
- Seamlessly Integrate AI into the eCommerce Workflow
Integrate those solutions in your company
Case Study Details
Approach
Image Quality Detection Using Computer Vision
- Addepto developed a system capable of automatically detecting and evaluating image quality, targeting common issues like blurriness, poor cropping, and distortion.
Leveraging State-of-the-Art Algorithms (VGG)
- The solution was built using the VGG model, a state-of-the-art convolutional neural network known for strong performance in image recognition tasks.
Training on Real User-Uploaded Data
- To ensure high accuracy, the model was trained on actual images uploaded to the Teezily platform, allowing it to learn platform-specific quality issues and standards.
Incorporating Predefined Quality Control Rules
- Beyond AI-driven detection, the solution was equipped with upfront rule-based checks to enforce specific quality thresholds and flag unacceptable content early.
Integration-Ready Software for eCommerce Backend
- The final product was delivered as a ready-to-integrate software module, enabling seamless connection with Teezily’s backend system for real-time image validation.
Our Team Expert Opinion
Teezily had no in-house AI team, but the company was deeply aware of AI potential, so the collaboration was very smooth. We decided to harness the sota computer vision algorithms and build an AI engine able to be incorporated into the Teezily eCommerce system.
Wojciech Drężek Data Scientist – Addepto
