book

Building AI Agents with LLMs, RAG, and Knowledge Graphs

by Salvatore Raieli, Gabriele Iuculano

July 2025

Intermediate to advanced

566 pages

16h 27m

English

Packt Publishing

Read now

Unlock full access

About the authors
Who this book is forWhat this book coversTo get the most out of this bookDownload the example code filesConventions usedGet in touchShare Your ThoughtsDownload a free PDF copy of this book
Technical requirementsRepresenting text for AIOne-hot encodingBag-of-wordsTF-IDFEmbedding, application, and representationWord2vecA notion of similarity for textProperties of embeddingsRNNs, LSTMs, GRUs, and CNNs for textRNNsLSTMsGRUsCNNs for textPerforming sentiment analysis with embedding and deep learningSummary
Technical requirementsExploring attention and self-attentionIntroducing the transformer modelTraining a transformerExploring masked language modelingVisualizing internal mechanismsApplying a transformerSummary
Technical requirementsDiscovering the evolution of LLMsThe scaling lawEmergent propertiesContext lengthMixture of expertsInstruction tuning, fine-tuning, and alignmentExploring smaller and more efficient LLMsExploring multimodal modelsUnderstanding hallucinations and ethical and legal issuesPrompt engineeringSummaryFurther reading
Technical requirementsUnderstanding the brain, perception, and action paradigmThe brainThe perceptionActionClassifying AI agentsUnderstanding the abilities of single-agent and multiple-agent systemsExploring the principal librariesLangChainHaystackLlamaIndexSemantic KernelAutoGenChoosing an LLM agent frameworkCreating an agent to search the webSummaryFurther reading

Technical requirementsExploring naïve RAGRetrieval, optimization, and augmentationChunking strategiesEmbedding strategiesEmbedding databasesEvaluating the outputComparison between RAG and fine-tuningUsing RAG to build a movie recommendation agentSummaryFurther reading
Technical requirementsDiscussing naïve RAG issuesExploring the advanced RAG pipelineHierarchical indexingHypothetical questions and HyDEContext enrichmentQuery transformationKeyword-based search and hybrid searchQuery routingRerankingResponse optimizationModular RAG and its integration with other systemsTraining and training-free approachesImplementing an advanced RAG pipelineUnderstanding the scalability and performance of RAGData scalability, storage, and preprocessingParallel processingSecurity and privacyOpen questions and future perspectivesSummaryFurther reading
Technical requirementsIntroduction to knowledge graphsA formal definition of graphs and knowledge graphsTaxonomies and ontologiesCreating a knowledge graph with your LLMKnowledge creationCreating a knowledge graph with an LLMKnowledge assessmentKnowledge cleaningKnowledge enrichmentKnowledge hosting and deploymentRetrieving information with a knowledge graph and an LLMGraph-based indexingGraph-guided retrievalGraphRAG applicationsUnderstanding graph reasoningKnowledge graph embeddingsGraph neural networksLLMs reasoning on knowledge graphsOngoing challenges in knowledge graphs and GraphRAGSummaryFurther reading
Technical requirementsIntroduction to reinforcement learningThe multi-armed bandit problemMarkov decision processesDeep reinforcement learningModel-free versus model-based approachesOn-policy versus off-policy methodsExploring deep RL in detailChallenges and future direction for deep RLLearning how to play a video game with reinforcement learningLLM interactions with RL modelsRL-enhanced LLMsLLM-enhanced RLKey takeawaysSummaryFurther reading
Technical requirementsIntroduction to autonomous agentsToolformerHuggingGPTChemCrowSwiftDossierChemAgentMulti-agent for lawMulti-agent for healthcare applicationsWorking with HuggingGPTUsing HuggingGPT locallyUsing HuggingGPT on the webMulti-agent systemSaaS, MaaS, DaaS, and RaaSSoftware as a Service (SaaS)Model as a Service (MaaS)Data as a Service (DaaS)Results as a Service (RaaS)A comparison of the different paradigmsSummaryFurther reading
Technical requirementsIntroduction to StreamlitStarting with StreamlitCaching the resultsDeveloping our frontend with StreamlitAdding the text elementsInserting images in a Streamlit appCreating a dynamic appCreating an application with Streamlit and AI agentsMachine learning operations and LLM operationsModel developmentModel trainingModel testingInference optimizationHandling errors in productionSecurity considerations for productionAsynchronous programmingasyncioAsynchronous programming and MLDockerKubernetesDocker with MLSummaryFurther reading
AI agents in healthcareBiomedical AI agentsAI agents in other sectorsPhysical agentsLLM agents for gamingWeb agentsChallenges and open questionsChallenges in human-agent communicationNo clear superiority of multi-agentsLimits of reasoningCreativity in LLMMechanistic interpretabilityThe road to artificial general intelligenceEthical questionsSummaryFurther reading
Other Books You May EnjoyPackt is searching for authors like youShare Your ThoughtsDownload a free PDF copy of this book

Content preview from Building AI Agents with LLMs, RAG, and Knowledge Graphs

10 Building an AI Agent Application

In the previous chapter, we discussed how an LLM can extend its capabilities by using other tools. We also saw some examples of how the use of multiple agents at the same time (instead of one) can be used to solve more complex tasks. We extensively discussed how these approaches can be used in various industries and how they can be revolutionary for so many applications. However, we also highlighted two of the limitations of agents: scalability and the complexity of connecting an agent with different tools.

In this chapter, we will expand on these challenges and show how we can overcome them. We will pick up from these two limitations. So far, we have treated multi-agent systems as standalone entities running ...