What this book covers

Chapter 1, Analyzing Text Data with Deep Learning, introduces how to process and represent natural language in a format suitable for machine learning models. It covers various text encoding techniques, from basic one-hot encoding and bag of words to more advanced representations such as TF-IDF and word2vec. The chapter then explores key deep learning architectures for sequential data, such as RNNs, LSTMs, GRUs, and CNNs, and demonstrates how to apply them to text classification tasks. By the end of this chapter you will understand how these foundations enable modern language models such as ChatGPT.

Chapter 2, The Transformer: The Model Behind the Modern AI Revolution, introduces attention mechanisms and explains how they evolved into the transformer architecture. The chapter highlights the limitations of earlier models such as RNNs and LSTMs, and shows how transformers overcame them to become the foundation of modern NLP. Key topics include self-attention, masked language modeling, training techniques, and internal model visualization. The chapter concludes by demonstrating real-world applications and laying the groundwork for understanding today’s LLMs.

Chapter 3, Exploring LLMs as a Powerful AI Engine, examines how the large-scale training of transformer models gave rise to today’s LLMs. The chapter explores their evolution, capabilities, and limitations, including techniques such as instruction tuning, fine-tuning, and alignment. It also introduces more compact and efficient LLM variants, multimodal models that handle multiple data types, and understanding challenges such as hallucinations, ethical concerns, and prompt engineering.

Chapter 4, Building a Web Scraping Agent with an LLM, introduces the concept of AI agents as an extension of LLMs, aimed at overcoming their ability to perform actions. The chapter explores the key characteristics of agents, and distinctions between single and multi-agent systems. It also presents the main libraries used for building agents and guides you through the creation of a web-scraping agent capable of retrieving information from the internet.

Chapter 5, Extending Your Agent with RAG to Prevent Hallucinations, explores how RAG could overcome key limitations of LLMs, such as outdated knowledge and hallucinations. The chapter explains how RAG enables an LLM to access external information sources through embedding and vector databases, thereby improving accuracy and adaptability. It also compares RAG with fine-tuning and demonstrates its practical use by building a movie recommendation agent.

Chapter 6, Advanced RAG Techniques for Information Retrieval and Augmentation, expands on the basic RAG architecture by introducing enhancements at every stage of the pipeline—data ingestion, indexing, retrieval, and generation. The chapter explores modular RAG, techniques for scaling systems with large datasets and user bases, and key concerns such as robustness and privacy. It also highlights current challenges and open questions surrounding the future development of RAG-based systems.

Chapter 7, Creating and Connecting a Knowledge Graph to an AI Agent, explores how to structure textual knowledge into knowledge graphs (KGs) to enhance information retrieval and reasoning in AI agents. The chapter introduces the concept of GraphRAG, where KGs are used to augment LLMs with structured contextual data. It covers how LLMs can be used to build KGs by extracting entities and relationships, how to use graphs for querying and reasoning, and discusses the benefits, limitations, and future directions of combining different approaches.

Chapter 8, Reinforcement Learning and AI Agents, explores how agents can learn by interacting with dynamic environments, adjusting their behavior based on experience. It introduces the fundamentals of reinforcement learning, explains how agents make decisions and improve over time, and demonstrates how neural networks can be used to guide behavior. The chapter concludes by discussing how LLMs can be combined with reinforcement learning to build more capable AI systems.

Chapter 9, Creating Single- and Multi-Agent Systems, explores how LLMs can be extended with tools and other models to form autonomous agents. It introduces the concept of single-agent and multi-agent systems, shows how LLMs can interact with APIs or external models, and presents key examples such as HuggingGPT. The chapter also covers agent coordination strategies, real-world applications in complex domains, and emerging business paradigms such as SaaS, MaaS, DaaS, and RaaS.

Chapter 10, Building an AI Agent Application, addresses the challenges of scaling and deploying AI agents in real-world applications. It introduces Streamlit as a rapid prototyping framework to create both frontend and backend components of an agent-based system. The chapter also covers key operational aspects such as asynchronous programming, containerization with Docker, and best practices for building scalable, production-ready AI solutions.

Chapter 11, The Future Ahead, explores the transformative potential of AI agents across industries such as healthcare and beyond. Building on the advancements discussed in earlier chapters, it reflects on the remaining technical and ethical challenges facing LLMs and agent systems. The chapter concludes by examining open questions and future directions in the development and deployment of intelligent AI agents.