Welcome to Multi-vector Image Retrieval, built in partnership with Qdrant. In this course, you'll learn how multi-vector image retrieval works and best practices for using it in your AI applications. For a long time, it was easiest to give LLMs access to only text data, but that leaves a lot of useful information inaccessible because it's stored in images and other forms of multimodal data. This course focuses on a collection of image retrieval techniques called Multi-vector. In contrast, standard vector retrieval works by representing each document or image with a single large vector. An image is retrieved if their vector is a good match for a query. Multi-vector techniques use many smaller vectors to represent the meaning of, say, smaller patches of an image. storing this detailed vector information then allows for fine-grained matching between the tokens in the text query and the image patches that make up your image or document. This approach to provide high quality image search that works well even on complex documents that combine text, images, slides, and storing all this vector data does lead to challenges with memory usage and search. But some new techniques address these problems and you'll explore these solutions as well in this course. To help you to understand these Multi-vector image retrieval techniques, I'm delighted to introduce your instructor for this course, Kacper Łukawski, who is an experienced AI engineer and a developer advocate for the vector database company Qdrant. Thanks, Andrew. You will first learn the fundamentals of multi-vector retrieval by seeing how it works for text. Then, you will see how this approach is adapted for image retrieval by a model called ColPali, and some techniques for optimizing its performance. Finally, you will explore how an approach called MUVERA aims to combine the best qualities of multi-vector and single-vector image retrieval. You will learn the foundational concepts underlying each method. and then we'll implement them inside of a functioning multimodal RAG system. You can expect a nice blend of conceptual and practical topics throughout this course. There's been a lot of excitement around the industry about image retrieval broadly and specifically multi-vector techniques. Most of the LLMs that people are using today are actually VLMs or vision language models. They are just as capable of processing images as text. The challenge has been building system that can accurately and efficiently retrieve image data to take advantage of that capability. The techniques in this course, like ColPali or MUVERA, are finally doing that. I really think this is the future of this field. and that will see it powering many exciting applications in the coming years. Many people have worked to create this course. From DeepLearning.AI, GT Wrobel contributed to its development. In the first lesson, you get familiar with multi-vector retrieval by getting hands on with ColBERT, which is a widely used multi-vector approach for text retrieval. Join Kacper in the next video, and let's get started.