VisiSense: An AI-Powered Visual Product Intelligence Platform, Redefining the Retail Product Catalog Generation Process

VisiSense is an open-source AI visual product analysis platform designed to address the pain points of time-consuming and highly repetitive e-commerce product catalog creation. It leverages multimodal visual large models to automatically convert product images into structured retail catalog content, supporting multiple LLM providers, real-time SEO scoring, and interactive chat Q&A. Adopting a FastAPI+React microservice architecture, it provides e-commerce teams with a complete product content generation solution.

In e-commerce operations, product catalog creation is time-consuming and it's hard to ensure quality consistency. VisiSense is developed by the cld2labs team and positioned as an AI visual product intelligence platform for retail product operation teams. Its core innovation is that users upload 1-5 product images, and the system automatically analyzes visual features to generate complete product data including titles, descriptions, attributes, etc., and provides real-time SEO scoring. This meets the demand for automated content generation in e-commerce and lowers the threshold for product listing.

VisiSense adopts a microservice architecture with separate front-end and back-end. The back-end is based on FastAPI, including VLM Service (coordinating visual analysis), Chat Service (dialogue interaction), Vision Client (multi-LLM adaptation), SEO Scorer (scoring optimization), Confidence Scorer (confidence evaluation), and Session Store (session caching). The front-end uses React18+TypeScript+Vite+Tailwind CSS, supporting drag-and-drop upload, real-time status display, etc. Deployment methods include one-click deployment via Docker Compose and local development mode.

1. Intelligent Image Analysis: Extract multi-dimensional attributes such as category and material from images, generate product identity, SEO content, attributes, selling points, keywords, and SKU suggestions; 2. Real-time SEO Evaluation: 0-100% scoring and grading, identify optimization points and support quick fixes/auto-enhancement; 3. Interactive Q&A: Provide context-aware answers based on analysis data to help understand product features.

VisiSense flexibly supports multiple LLMs: OpenAI (GPT-4o, high quality, suitable for production), Groq (fast inference, suitable for testing), Ollama (local deployment, privacy protection), OpenRouter (multi-model switching), and custom API endpoints (compatible with OpenAI format), meeting different cost, privacy, and performance needs.

Applicable to e-commerce operation scenarios such as bulk product listing (shortening new product cycles), multilingual content localization (combined with translation APIs), supplier product information standardization (unified format), and marketing content inspiration generation (obtaining selling point materials via Q&A).

Usage Notes: AI-generated content requires manual review; depends on image quality (sufficient light, high resolution); sessions expire after 30 minutes of inactivity, so export in time; cloud solutions (e.g., OpenAI) require consideration of token costs for large-scale use.

VisiSense is an excellent practice of applying multimodal AI to business, freeing human creativity to focus on strategic ideas. Its values include efficiency improvement (second-level generation), consistent quality, SEO-friendliness, and flexible deployment. In the future, with the evolution of VLMs, it will be implemented in more vertical fields, providing e-commerce teams with a starting point for open-source exploration.