TileLens - Visual Search Engine

TileLens is an AI-powered visual search engine built for tile manufacturers and retailers to simplify product discovery using images instead of traditional keyword-based search. Users can upload a tile image, and the system intelligently finds visually similar designs by analyzing patterns, textures, colors, and surface details.

The platform uses deep learning–based embedding models to convert tile images into high-dimensional vector representations. These embeddings are stored and indexed using PostgreSQL with pgvector, enabling fast and accurate similarity search across thousands of tile designs in real time.

The application was designed with scalability and production-readiness in mind. Backend services were containerized using Docker and deployed on a Kubernetes (k3s) cluster, ensuring reliable performance, simplified deployments, and efficient resource management for handling large-scale image processing workflows.

• Built an image-based visual search engine that allows users to search tile designs using uploaded images instead of text queries
• Processed and indexed 15,000+ tile images using vector embeddings for high-accuracy visual similarity matching
• Implemented deep learning embedding pipelines to extract visual features such as texture, color patterns, and design structure
• Integrated PostgreSQL with pgvector to perform efficient high-dimensional vector similarity searches
• Designed and developed REST APIs using Flask for image upload, embedding generation, and similarity search operations
• Optimized nearest-neighbor search performance to deliver fast and real-time search recommendations
• Containerized backend services using Docker for consistent development and deployment environments
• Deployed the complete system on a Kubernetes (k3s) cluster to improve scalability, reliability, and service orchestration
• Structured the application using modular backend services for easier maintenance and future feature expansion
• Focused on building a clean and responsive user experience for image upload, result visualization, and search interactions

System Architecture

Architecture includes image upload API, embedding service, vector database, and similarity search engine.

Search Workflow

User uploads an image → deep learning model generates embedding → pgvector performs similarity search → top matching tile designs are returned.

Application Screens

User interface for image upload, similarity results display, and real-time search experience.