Text + Vision AI Search Platform
Image and text-based product discovery for a modern marketplace
An AI-powered search platform for e-commerce (marketplaces) supporting image-based and text-based product discovery using FastAPI, PostgreSQL, Qdrant, embeddings, and object detection.
Technology stack
Problem
Shoppers needed a way to find products by uploading a photo or describing them in natural language, not just by keyword matching against titles.
Solution
Built a retrieval pipeline that combines object detection with text and image embeddings, indexed in a vector database for fast approximate nearest-neighbor search across the catalog.
Technical contribution
Designed the FastAPI service architecture, the embedding and indexing pipeline, and the Qdrant collection schema, and integrated YOLO-based object detection for image queries.
Architecture
FastAPI service layer in front of a Qdrant vector store, with PostgreSQL for catalog and metadata, and a background pipeline that generates and refreshes embeddings as products change.
Performance considerations
Approximate nearest-neighbor indexing keeps query latency low even as the catalog grows into the hundreds of thousands of items.
Security considerations
Search and ingestion endpoints are isolated behind service-level authentication, with strict input validation on all uploaded media.
Deployment model
Containerized with Docker and deployed as independently scalable services for the API layer, the vector store, and the embedding workers.
Outcome
Shoppers can now search by photo or free-text description, with relevant results returned from a fully automated indexing pipeline.