Imagine an AI assistant that can not only read but also *see*—an assistant capable of understanding complex diagrams in a manual or identifying a faulty component in a machine from a photo. That's the potential of multimodal Retrieval Augmented Generation (RAG), a technique that's taking AI beyond text to a new level of understanding. Large Language Models (LLMs) are great at generating text, but they often struggle with real-world knowledge and can even 'hallucinate' incorrect information. RAG systems help resolve this issue by allowing LLMs to pull information from external sources like manuals. But what if those manuals rely heavily on visuals, such as in industrial settings? That's where multimodal RAG comes in. By adding images to the mix, researchers are seeing significant improvements in the accuracy and relevance of AI-generated answers. In a recent study, researchers explored how to best incorporate images into industrial RAG systems. They tested two methods: one used multimodal embeddings to link images and text, while the other converted images into text summaries. Interestingly, both methods showed promising results, offering comparable performance. However, generating text summaries from images offered more flexibility for tailoring and optimizing the system. The biggest challenge? Image retrieval. While current LLMs excel at processing text, getting them to 'see' and understand the right image from a vast collection remains tricky. Imagine having to find a specific diagram in a thousand-page manual – now ask an AI to do that! This research highlights the exciting potential of multimodal AI assistants, especially in industries like manufacturing, engineering, and maintenance, where visual information is crucial. It also points to the next frontier: improving image retrieval and creating robust, domain-specific datasets to train these powerful systems.