Building Multimodal Search and RAG

Building Multimodal Search and RAG is a free, roughly 1.5-hour DeepLearning.AI short course built with Weaviate and taught by Sebastian Witalec, Weaviate's Head of Developer Relations; it is a strong, hands-on intro to any-to-any (text, image, video) retrieval and multimodal RAG for engineers who already know basic Python and standard text RAG. Across eight short video lessons and six runnable notebooks, you implement contrastive multimodal embeddings on MNIST, build text-to-any and any-to-any search with Weaviate plus Google Vertex AI's multimodalembedding model, query images with Gemini Vision, assemble an end-to-end multimodal RAG pipeline, extract structured data from invoices and tables, and build a multi-vector recommender on a movies dataset. It holds a real 4.5/5 rating from 43 reviews on Coursera (about 6,730 enrolled there), and learners praise it as a concise, practical primer rather than a deep theoretical course. The major caveat is currency: as of 2026 the course has been flagged for maintenance and community threads report broken lab code tied to Google's deprecation of the PaLM API and Vertex AI authentication headers, so the notebooks may not run as-recorded without fixes. Treat it as a conceptual and pattern walkthrough worth taking for free, but expect to debug or substitute the Google/Vertex API calls if you run the labs end-to-end.

The content, instructor, and free price make this a high-value 1.5-hour primer on multimodal embeddings and RAG, but it is short, certificate-less, and (as of 2026) its labs have known breakage from deprecated Google PaLM/Vertex APIs, so it is a clear take only for learners comfortable patching outdated API code.

Best for: Intermediate ML/AI engineers and backend developers who already understand standard text-based RAG, embeddings, and vector databases, and want a fast, hands-on introduction to extending those patterns to images and video using multimodal embeddings, Weaviate, and Gemini/Vertex. Ideal for people evaluating whether multimodal search/RAG fits a product, or who learn best from short runnable notebooks over long lectures.

Skip if: Complete beginners with no Python or no prior RAG/embeddings exposure (the course is explicitly intermediate and moves quickly); anyone who needs a completion certificate; learners who want deep theory on multimodal model training; and anyone who needs labs that run flawlessly out of the box right now, given the documented deprecated-API breakage and maintenance status.