AI Glossary

A personal glossary of AI concepts, platforms and products.

This glossary is a personal reference and explainer covering common terms related to Artificial Intelligence (AI). It includes concepts, platforms, frameworks, metrics, and popular models or products. It’s not intended to be definitive or exhaustive — the field moves fast — but it captures many of the key ideas and names that come up frequently in current AI discussions. All terms are listed in alphabetical order for easy reference.

Agent / AI Agent / Agentic AI

A system capable of making decisions and taking actions autonomously or semi-autonomously, often by planning and using tools or APIs. Agentic AI refers to models that behave like agents, completing multi-step tasks with minimal input.

AI Ready

Describes whether a system, product or organization is technically and structurally prepared to integrate AI tools. This could include having APIs or structured data available, being modular or extensible, having workflows that can be automated or enhanced with AI, or a culture open to using AI-driven systems.

API (Application Programming Interface)

A set of rules and conventions provided by a specific system or service that defines how other software can interact with it programmatically. Each API is custom to the system it exposes and acts like a machine-readable interface — the software equivalent of a user interface. APIs are used by developers to send and receive data, trigger actions, or integrate systems. They typically require writing code that follows the API’s structure and protocols. With millions of APIs in use, they are the standard method for connecting to services across the web, including AI models, databases, cloud platforms, and apps.

Artificial General Intelligence (AGI)

A theoretical form of AI with human-like general reasoning and problem-solving capabilities across diverse tasks. Not yet achieved.

Artificial Intelligence (AI)

A broad term referring to machines or software that simulate human-like intelligence, including learning, reasoning, perception, and language understanding.

Auto-GPT

An experimental open-source project that chains LLM calls together to autonomously plan and execute complex tasks.

AX (AI Experience)

A term used to refer to the quality of the human user’s interaction with an AI system — including factors like responsiveness, clarity, trustworthiness, helpfulness, and tone. Just as UX (User Experience) focuses on how users interact with software, AX emphasizes the design and feel of AI-driven tools and assistants.

BabyAGI

A minimal, open-source AI agent framework designed to perform recursive task management using LLMs and memory.

BLEU / ROUGE / F1

Metrics used to evaluate the quality of natural language generation, often in machine translation or summarization tasks.

Chatbot

A software application that uses natural language to simulate human conversation, often using rules or AI models like LLMs. These are typically used to handle customer support queries.

ChatGPT (OpenAI)

A conversational AI based on OpenAI’s GPT models. ChatGPT is widely used for interactive tasks, from Q&A to writing.

Claude (Anthropic)

A family of large language models developed by Anthropic, known for their safety-focused alignment techniques.

Context Engineering

The practice of designing and structuring the context passed into a language model — including instructions, examples, and memory — to influence its behavior and improve output. An extension of prompt engineering, especially important in multi-step and agent workflows.

CoPilot (Microsoft)

A suite of AI assistants integrated into Microsoft tools like Word, Excel, and GitHub, powered by OpenAI models.

DALL·E (OpenAI)

An image generation model that creates pictures from text prompts. It uses diffusion and transformer techniques.

Deep Learning

A subset of machine learning involving neural networks with multiple layers. It powers most modern AI models.

Diffusion Model

A generative model that creates data (like images) by reversing a process of gradual noise addition, used in tools like DALL·E and Midjourney.

Docling

An open-source project for extracting structured data from unstructured documents like PDFs, DOCX, or HTML files. Designed to make real-world documents machine-readable, which is a frequent challenge in AI-driven workflows and integrations.

Embedding / Vector Embedding

A mathematical representation of data (like words, images, or documents) in a high-dimensional space, used for similarity search and semantic understanding.

Few-shot Learning

A method where models perform tasks with only a few examples in the prompt. Common with LLMs.

Fine-tuning

The process of adapting a pre-trained model to a specific dataset or task, improving performance for narrow applications.

Foundation Model

A large, pre-trained model trained on broad data and capable of being adapted to many tasks, e.g. GPT-4, Claude.

Generative AI

A category of AI focused on generating new content — text, code, images, audio, or video — from learned patterns.

Gemini (Google)

Google DeepMind’s family of AI models, succeeding Bard. Designed for multimodal use and integration across Google’s ecosystem.

GPT (Generative Pretrained Transformer)

A family of large language models developed by OpenAI, based on the transformer architecture. GPT models are trained on massive amounts of text data to predict and generate human-like language. GPT models are widely used in AI assistants, chatbots, creative writing tools, coding helpers, and business automation.

GPTs (Custom GPTs in ChatGPT)

Custom AI assistants built using the ChatGPT app. These “GPTs” are built on top of the base GPT-4 (or other models), but include additional context, instructions, personality traits, files, APIs, or tools defined by the creator. They behave like tailored versions of ChatGPT designed for specific tasks, e.g. a legal assistant, writing coach, or code reviewer.

Hallucination

When an AI generates plausible-sounding but factually incorrect or fabricated information.

Hugging Face

A platform and hub for open-source machine learning models and tools. Hosts thousands of models and datasets.

Inference

The process of running an AI model on input data to generate output. Also refers to the infrastructure that serves model predictions.

Inference Endpoint

An API or service that exposes a model to receive inputs and return predictions or generations.

Langchain

A popular open-source framework for building applications powered by language models. Langchain allows developers to combine LLMs with external tools, APIs, memory, and documents — enabling complex workflows like question answering over PDFs, data extraction, or multi-step decision-making. It’s especially useful for building RAG pipelines, agent-based systems, and AI assistants that go beyond single prompts.

LangGraph

A companion framework to Langchain for building stateful, graph-based LLM applications. It lets developers define agent behaviors, memory, and branching logic as part of a persistent, multi-step graph. LangGraph is often used for structured agents, task planners, and long-running AI workflows — such as customer support flows, research assistants, or autonomous agents that call tools and APIs.

Large Language Model (LLM)

A type of AI trained on massive amounts of text to generate and understand natural language. Examples: GPT-4, Claude, Gemini.

Latency / Throughput

Performance metrics: latency measures response time per request; throughput measures total requests handled over time.

LLaMA (Meta)

A family of open-weight LLMs developed by Meta, widely used in the open-source AI community. A community platform for sharing AI-powered apps and agents. Often features playful, experimental, or helpful tools created with LLMs.

LlamaIndex (formerly GPT Index)

A framework for building custom search and retrieval systems on top of LLMs. Often used in Retrieval-Augmented Generation (RAG) pipelines to index documents and feed relevant context into prompts.

Lovable

A community platform for creating, sharing, and remixing lightweight AI apps and agents, often built with LLMs. Focused on experimentation and creativity, Lovable encourages a “vibe coding” approach — building by intuition and play rather than formal engineering. It’s a popular hangout for developers, tinkerers, and creators working at the intersection of AI and user experience.

Machine Learning (ML)

A subfield of AI where systems learn from data to improve performance on specific tasks without explicit programming.

Midjourney

A popular generative image model hosted via Discord, known for its distinctive and artistic outputs.

Model Context Protocol (MCP)

A proposed open standard for managing the flow of ‘context’ between users, apps and AI models. MCP provides a structured way to pass messages, roles, memory and metadata into LLMs — enabling consistent, rich interactions across apps, APIs and systems. It separates the model’s understanding from the UI or backend logic. MCP has the potential to transform how AI systems are integrated into products and workflows by making context management portable, inspectable, and shareable across tools.

Mistral

A family of open-weight LLMs known for efficiency and performance, commonly used in privacy-focused or self-hosted deployments.

Multi-Agent System (MAS)

A system where multiple AI agents interact or collaborate to solve tasks, often with coordination and communication.

Natural Language Processing (NLP)

A branch of AI focused on understanding, generating and interpreting human language.

Neural Net / Neural Network

A computational model inspired by biological brains, used as the basis for deep learning systems.

OpenAI

The AI research company behind GPT, ChatGPT, DALL·E, and Sora. Known for pushing the boundaries of generative AI.

Perplexity (metric)

A measurement of how well a language model predicts a sample. Lower perplexity typically indicates better performance.

Perplexity (AI tool)

A conversational search engine and assistant that integrates LLMs with real-time search results.

Planning / Tool Use

The process where an AI (especially agents) breaks tasks into steps and uses APIs or tools to complete them.

Prompt Engineering

The craft of designing effective inputs for LLMs to elicit the desired outputs.

Prompt Template

A reusable prompt structure with variables that can be filled programmatically for consistent behavior.

RAG (Retrieval-Augmented Generation)

A technique where an LLM is augmented with external context (e.g. from documents or vector databases) at runtime.

Sora (OpenAI)

A generative video model developed by OpenAI, capable of creating short videos from text prompts.

Temperature / Top-k / Top-p

Parameters controlling randomness in generation. Lower temperature = more deterministic; top-k and top-p limit token choices.

Token / Tokenization

A unit of text used by LLMs for processing. Tokenization breaks input into these units before inference.

Transformer

A neural network architecture that enables models like GPT to understand language via attention mechanisms.

Vector Database

A special kind of database designed to store and search embeddings (numerical representations of text, images, or documents) — often used in RAG systems. Examples include Pinecone, FAISS, and Weaviate.

Vector Embedding / Embedding

A way of turning data like text or images into a list of numbers, so that AI systems can compare them for similarity. Often used for search or memory retrieval.

VEO / VEO-3

A family of advanced multimodal AI models developed by Google DeepMind. VEO-3, for example, is capable of understanding and generating text, images, audio, and video — though it is not yet widely released. VEO is a term sometimes used for Video-Enhanced Output or tools like OpenAI’s Sora, depending on the context.

Vibe Coding

A loose approach to creating code with AI tools — often driven more by intuition, playfulness, or creative exploration than formal design. Common in early prototyping or when “coding by feel” with LLMs and no-code tools.

Zero-shot Learning

When a model performs a task it hasn’t explicitly been trained for, using general capabilities.