Traditional image galleries rely on manual tags or basic metadata, which do not scale and fail to capture the semantic meaning of visual content. This project addresses that limitation by building a full stack AI powered image gallery that enables semantic search, automated tagging, and conversational interaction over images.
The system combines a React and TypeScript frontend with a Node.js and Express backend that performs on device inference using the CLIP model via Transformers.js. Images are fetched from the Picsum Photos API and enriched in real time with embeddings, category labels, and contextual descriptions, allowing the system to understand visual content beyond keywords.
User queries from both search and chat interfaces are converted into vector embeddings and matched against image representations using similarity scoring. This enables natural language search such as describing a scene instead of relying on exact matches, while a caching strategy improves performance by reducing repeated model computations.
The application also includes an AI driven chat assistant that interprets user intent and dynamically triggers search and filtering actions. The overall architecture separates frontend, API, and AI services, ensuring asynchronous processing and a responsive user experience while demonstrating practical implementation of multimodal AI and scalable system design.