Insights from Diverse Projects in Image Search and RAG

Explore practical insights and real-world applications in a technical talk focusing on vector search implementations across four distinct projects. Delve into the complexities of image search engines, including matching clothes and objects in videos to purchasable items, and deduplicating real estate listings with similar property images. Learn about Retrieval-Augmented Generation (RAG) applications in both internal company documentation and commercial medical contexts. Examine the performance characteristics and limitations of key models like DinoV2 and Ada-2, while understanding the practical implementations using vector databases such as Qdrant and Milvus. Gain valuable insights from Lead Data Scientist Noé Achache's extensive experience in computer vision, structured data prediction, and Large Language Models (LLMs), with detailed discussions on model selection, testing methodologies, and data access considerations for various use cases.