Andrew Ng and Matt Robinson from Unstructured teach how to prepare PDFs, Word docs, PowerPoint, and other files for RAG systems. The course covers extracting tables, analyzing document images with vision transformers, and preserving structure and metadata so LLMs can reason over the content better.
I'm delighted to introduce Pre Processing unstructured data for LLM applications. Built in partnership with Unstructured and taught by Matt Robinson, Retrieval Augmented Generation or RAG is increasingly adopted by enterprises. In this course you learn techniques for getting many types of unstructured data like text, images, tables from many different data sources like PDF, PowerPoint, Word and so on in a way that lets your LM RAG system access all of this information, thus enabling to generate much better answers when building an LLM RAG system. One component that has a huge impact on the result quality but is often overlooked is the work needed to get that data to the LLM. Serecomplete has a lot of different file types and data formats like numeric data and Excel spreadsheets reports in Word or PDF presentations in PowerPoint HTML pages. How do you get an interact system to use all of these different types of file formats? Further, a PDF might itself have tables and images inside that PDF and a PowerPoint might also contain tables and images. Parsing these different formats and normalizing the data, for example making sure an image, whether from a PDF or a PowerPoint is represented similarly will help your LLM better retrieve and use all this information. In addition to normalization, it also helps to keep the high level structure of data from the original document by preserving metadata that tells you, for example, where a certain table or image had come from in that larger document and also keep track of things like what was the title subheading and so on within that larger document. This type of metadata helps your ELM RAG system do better retrieving as well as reasoning over that data. So as you can see, there are a lot of important practical details for creating a RAG system from different data sources. I'm delighted to introduce the instructor Matt Robinson, who is head of Product at Unstructured, which provides a rich set of tools for OM data pre processing. Matt has helped many developers build OM applications that use and combine data from diverse sources.
Thanks Andrew, I'm really excited to be teaching this course. In this course you'll learn about data pre processing for LLM application development, focusing on how to work with different document types like PDFs, PowerPoints, Word docs and HTML. You'll learn to extract and normalize content and enrich it with metadata to improve search results. This course covers techniques for document image analysis including document layout, detection and vision transformers, where you'll learn how to extract and understand tables. After learning these concepts, you'll put them all together to build a ragbot out of a Corpus that includes PDF, PowerPoint and Markdown documents.
This course teaches important techniques that have a huge practical impact on how well your rag system will perform. Please enjoy the course.
