What you'll learn

Learn to extract and normalize content from a wide variety of document types, such as PDFs, PowerPoints, Word, and HTML files, tables, and images to expand the information accessible to your LLM.
Enrich your content with metadata, enhancing retrieval augmented generation (RAG) results and supporting more nuanced search capabilities.
Explore document image analysis techniques like layout detection and vision and table transformers, and learn how to apply these methods to preprocess PDFs, images, and tables.

Course recap

PRO

This course teaches techniques for preprocessing diverse document types (HTML, PDFs, PowerPoints, images, Markdown) into normalized elements suitable for LLM applications, particularly RAG. It covers both rules-based parsing and model-based approaches like document layout detection (YOLOX) and vision transformers (Donut).

Concept map

Key concepts

Document layout detectionObject detection models (YOLOX) that draw bounding boxes around document elements, then extract text via OCR or direct extraction
Vision transformersModels like Donut that take document images as input and produce structured JSON output in a single step
Retrieval-Augmented GenerationThe end application: preprocessing documents to enable semantic search and LLM-powered Q&A
EmbeddingsOpenAI embeddings used to load preprocessed documents into ChromaDB for vector search
Metadata extractionParent-child relationships, section hierarchy, element types used for filtering and hybrid search
Chunkingchunk_by_title strategy that groups elements by section before splitting

Lesson highlights

1.Introduction — Overview of preprocessing needs for LLM applications
2.Overview of document preprocessing — Element types, rules-based vs. model-based parsing
3.Preprocessing HTML files — partition_html, tag-based element categorization
4.Preprocessing PDFs and Images — Document layout detection vs. vision transformers, YOLOX model, fast/hi_res strategies
5.Extracting tables from PDFs — Table extraction as HTML, handling complex PDF layouts
6.Metadata and chunking — Element metadata (parent_id, section), chunk_by_title, filtering techniques
7.Build Your Own RAG Bot — End-to-end RAG with PDF, PPTX, Markdown corpus; LangChain, ChromaDB, hybrid search
8.Conclusion

About this course

Enhancing a RAG system’s performance depends on efficiently processing diverse unstructured data sources.

In this course, you’ll learn techniques for representing all sorts of unstructured data, like text, images, and tables, from many different sources and implement them to extend your LLM RAG pipeline to include Excel, Word, PowerPoint, PDF, and EPUB files.

Join this course and learn:

How to preprocess data for your LLM application development, focusing on how to work with different document types.
How to extract and normalize various documents into a common JSON format and enrich it with metadata to improve search results.
Techniques for document image analysis, including layout detection and vision transformers, to extract and understand PDFs, images, and tables.
How to build a RAG bot that is able to ingest different documents like PDFs, PowerPoints, and Markdown files.

Apply the skills you’ll learn in this course to real-world scenarios, enhancing your RAG application and expanding its versatility.

Who should join?

Anyone who is interested in learning how to effectively process and use diverse data types and formats to build high-performing LLM RAG systems.

Preprocessing Unstructured Data for LLM Applications

What you'll learn

Course recap

Concept map

Key concepts

Lesson highlights

About this course

Who should join?

Course Outline

Elevate your learning experience with Pro

Instructor

Matt Robinson

Preprocessing Unstructured Data for LLM Applications

Want to learn more about Generative AI?