What you will learn in Grokking AI Applications
An illustrated guide for programmers and other curious people
Chapter by chapter
Concepts explained in this book
- retrieval-augmented generationA generative AI technique that retrieves relevant external information and supplies it to a model as context for generation.
- AI agentArtificial intelligence whose systems can operate autonomously, make decisions, and perform tasks with limited human intervention.
- language modelA large language model is a machine-learning model trained on large text collections to predict and generate sequences of natural-language tokens.
- vector databaseA vector store is a specialized database that stores text chunks or other items together with vector representations for similarity-based retrieval.
- similarity searchA search method that finds stored vectors whose representations are most similar to a query vector.
- promptA prompt is an instruction or input provided to a generative AI model to elicit a response or guide content generation.
- embeddingAn embedding is a numerical vector representation of a text unit that encodes information used for model computations.
- toolAn external callable capability that an AI application or agent can invoke to perform an operation or access information.
- text chunkA contiguous portion of text created by splitting a larger document into smaller units for processing or retrieval.
- split text componentA Langflow workflow component that divides source content into chunks according to configured sizing and overlap settings.
- templateA structured prompt format organized into role, task, features, and examples.
- data aggregationData aggregation is the process of combining multiple data records or values into summarized results.
- joinA relational data operation that combines rows from two tables using matching values in one or more key columns.
- JSONJSON is a text-based data-interchange format that represents structured values using objects, arrays, and primitive values.
- left outer joinA join that retains every row from the left table and adds matching rows from the right table when available.
- ComposioComposio is a software platform that provides integrations enabling AI agents to interact with external applications and services.
- MCP serverAn MCP server is a service that implements the Model Context Protocol to expose tools and related capabilities to compatible clients or agents.
- MCPThe Model Context Protocol is an open standard for discovering and invoking tools exposed by servers for use by AI applications and agents.
- LangFlowLangflow is an open-source visual platform for building applications and workflows around language models and related AI components.
- delimiterA delimiter string or pattern used to divide source text into pieces before they are merged or trimmed to fit chunking constraints.
- prompt engineeringPrompt engineering is the systematic design and refinement of prompts to influence a generative model's outputs.
- sequential agent structureAn agentic structure in which multiple agents execute in a fixed order, passing each agent's output to the next agent.
- fine-tuningFine-tuning is additional training of a pretrained model on more specific data or objectives to adapt its behavior.
- tokenA token is a discrete text unit that a language model converts into numerical representations and processes as input or output.
- GPT modelGPT is a generative pretrained transformer language model that predicts likely continuations of text.
- low-code platformA software development tool that enables applications to be built with limited hand-written code through visual components, configuration, or generated code.
- observabilityThe capability to understand an application's internal behavior from its outputs, telemetry, logs, traces, and related runtime signals.
- LangSmithLangSmith is a development and monitoring platform for tracing, evaluating, and debugging language-model applications.
- FAISSFAISS is a software library for efficient similarity search and clustering of dense vectors.
- contractA defined interface through which software programs invoke operations or exchange data.
- chat widgetA chat widget is an embeddable web interface that provides conversational interaction with a connected application or service.
- portA port is a typed connector on a workflow component through which data, models, variables, or other resources enter or leave the component.
- environment variableAn environment variable is a named configuration value supplied to a running process through its execution environment.
- MCP Tools componentThe MCP Tools component is a Langflow component that connects to an MCP server, retrieves its available tools, and exposes a selected tool for invocation.
- parser componentA workflow component that formats parsed document content and associated metadata into text for downstream processing.
- read file componentThe Read File component reads an uploaded file and outputs the text or parsed content it contains.
- agentic structureAn agentic structure is an organizational design in which multiple specialized agents collaborate under a coordinating agent with assigned roles and responsibilities.
- adversarial rolesA role-based prompting pattern in which distinct perspectives are assigned to produce critique, tension, or complementary reasoning.
- Chroma DBChroma is an open-source vector database for storing embeddings and associated data and performing similarity-based retrieval.
- flow editorA visual workspace for designing and orchestrating connected application components in Langflow.
- OpenAI PlaygroundA testing environment for interacting with an AI application and observing its results during development.
- toolsetA collection of tools together with structured descriptions of their capabilities and parameters.
- OpenAIEmbeddingsOpenAI Embeddings is an API capability that generates vector representations for supplied text.
- retrieve and answerRetrieve and answer is a workflow stage that searches stored document representations, combines retrieved text with a user query, and generates an answer with a language model.
- supervised learningSupervised learning is a machine learning approach that learns to predict outcomes from labeled examples.
- transformer architectureA transformer is a neural-network architecture for sequential data that uses attention to relate elements of an input sequence.
- attentionAn attention mechanism computes the relative relevance of input elements so a model can focus on the most informative parts of a sequence.
- backpropagationBackpropagation is an optimization procedure that propagates prediction error through a neural network to adjust its parameters.
- datasetA dataset is an organized collection of data examples used to train or evaluate a machine-learning model.
- pretrained modelA pretrained model is a model whose parameters have already been learned from an initial training corpus before later application or specialization.
- trainingModel training is the iterative optimization process that learns parameter values from examples to improve predictions.
- knowledge baseAn organized collection of information that can be searched or supplied to an application.
- information retrievalThe process of finding and obtaining relevant information from a data source for use in an AI application.
- instruction repetitionA prompting technique that repeats critical instructions within a prompt to reinforce their importance and clarify the desired response.
- KNIMEKNIME is a visual analytics platform for designing data science and machine-learning workflows.
- retrieved documentTextual information retrieved from a knowledge source and supplied to a language model alongside a user query.
- global variableA named value managed centrally so it can be reused across multiple Langflow flows.
- machine learningAn area of artificial intelligence in which machines learn patterns or behavior autonomously from data rather than following only fixed rules.
- Wikipedia APIA public interface for searching Wikipedia articles and retrieving their summaries, source URLs, and full text programmatically.
- deploymentDeployment is the process of making an AI application available to users or operational systems.
- hallucinationAn incorrect or nonsensical output generated by a language model, often because of insufficient context or vague instructions.
- prompt structureThe organization of a prompt into clearly distinguishable parts that a language model can parse.
- shot learningA prompting approach that provides zero or more examples to guide the output behavior of a language model.
- Hugging Face SpaceHugging Face Spaces is a platform for hosting, running, and publicly showcasing interactive machine-learning demos and applications.
- virtual machineA virtual machine is a software-defined computer that runs on shared physical infrastructure and provides an isolated environment for operating systems and applications.
- ChatGPTChatGPT is a conversational AI system built on generative language models and trained to respond to human users.
- context windowA context window is the maximum number of tokens a language model can process in one prompt.
- cloud computingCloud computing is the delivery of computing resources such as processing, memory, storage, and networking over the internet on a rented or usage-based basis.
- memoryApplication memory is persisted conversational or task-related information that a software system can retrieve across interactions.
- multi-agent systemAn AI system composed of multiple agents that collaborate or coordinate to accomplish tasks.
- endpointA network-accessible address exposed by an API for receiving requests to perform an operation or provide data.
- calculator toolA tool that evaluates arithmetic expressions on behalf of an AI agent and returns numeric results.
- agent orchestration protocolAn agent orchestration protocol is a defined communication or coordination interface that enables agents and tools to collaborate.
- contextContext is a numerical representation derived from the input text that summarizes information relevant to predicting a continuation.
- reinforcement learningReinforcement learning is a machine learning approach in which an agent learns through trial and error while interacting with an environment to achieve goals.
- reinforcement learning with human feedbackReinforcement learning from human feedback is a training technique that optimizes model behavior using human evaluations of its responses.
- unsupervised learningUnsupervised learning is a machine learning approach that discovers patterns or structure in data without labeled examples.
- chain-of-thought promptingA prompting technique that asks a language model to express intermediate reasoning steps before presenting a conclusion.
- self-validationA prompting technique that asks a model to verify, cross-check, or justify its own response before or while presenting it.
- uvA fast Python package manager and runner that also manages Python virtual environments.
- virtual environmentAn isolated Python environment containing a project-specific interpreter and installed packages.
- fallbackA prompt instruction that specifies an alternative response when the model cannot complete the requested task or find the requested information.
- decompositionA problem-solving technique that divides a complex request into smaller, ordered subtasks.
- nodeA KNIME node is a modular workflow unit that accepts configured inputs, performs a specific processing task, and produces outputs.
- FlowiseAn open-source visual tool for building applications and workflows based on language models.
- ETLExtract, transform, load is a data-integration process that extracts data from sources, transforms it, and loads it into a target system.
- KNIME WorkbenchKNIME Workbench is the desktop development interface in which users create and configure KNIME workflows.
- stateThe stored sequence of prior user inputs and model outputs that represents an ongoing conversation.
- visual flowA graphical representation of connected application components and the data or control paths between them.
- HyDEA query-transformation technique that generates hypothetical answer text and uses its embedding for document retrieval.
- query rewritingA technique that rephrases a user query to improve its semantic match with retrievable document content.
- data transformationData transformation is the modification, restructuring, cleaning, or reformatting of data to meet processing or storage requirements.
- self-hostingSelf-hosting is operating an application on infrastructure that an individual or organization owns or directly manages.
- subagentA specialized agent that performs a delegated part of a larger task under the coordination of another agent.
- run flow componentA Langflow component that exposes an entire flow as an invocable tool for another flow or agent.
- A2AThe Agent-to-Agent protocol is a standard for direct communication, task negotiation, and coordination between AI agents.
- ACPThe Agent Communication Protocol is an open protocol for standardizing agent messaging and enabling agents to be shared across projects and organizations.
- structured dataStructured data is information organized according to a predefined schema of fields, records, tables, or other machine-readable structures.
- dialogueA record of earlier messages exchanged between a user and an agent or model.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.