AI Lister: Automated Listing Tool

The AI Lister is a PyQt5-based desktop application developed for my Bachelor’s thesis to automate clothing listings on e-commerce platforms like eBay and Vinted. It integrates computer vision for color extraction, a fine-tuned BERT model for category classification, and a large language model (TinyLlama) for generating platform-specific titles and descriptions, streamlining the listing process for sellers.

Key Features

• Color Extraction: Uses SAM2 for segmentation and K-means clustering to extract dominant colors from clothing images, achieving a 99.3% segmentation success rate.
• Category Classification: Fine-tuned BERT model classifies items into platform-specific categories (e.g., 0.9964 accuracy for eBay, 0.9738 for Vinted).
• Text Generation: TinyLlama generates tailored titles and descriptions for eBay (keyword-rich) and Vinted (concise, trendy) using few-shot prompting.
• User Interface: Multi-page PyQt5 interface with image carousel, progress updates, and editable outputs for a seamless user experience.
• Custom Dataset: Built synthetic datasets for training and evaluation, addressing gaps in e-commerce data.

Resources

Explore the code, datasets, and Jupyter notebooks for reproducibility on the project’s GitHub repository: Automated Listing Tool.