The AI Glossary: Every Term Explained in Plain English
A reference page, not an essay. Every term defined in two sentences. Links to full explainers where available.
The fundamentals
LLM (Large Language Model) A type of AI trained on massive amounts of text that can generate human-like responses, answer questions, and reason about language. Examples include GPT-4, Claude, and GLM. The “large” refers to the number of parameters: billions or trillions of numbers that encode language patterns.
RAG (Retrieval-Augmented Generation) A technique where the AI searches your documents before answering, so responses are grounded in real data instead of training memory. It turns a language model from a know-it-all that guesses into a researcher that looks things up. Full explainer →
Vector Embeddings A way of converting text into arrays of numbers where proximity means similarity. Documents about the same topic land near each other in mathematical space, which is how semantic search works. Full explainer →
Fine-Tuning A training process that takes a pre-trained model and continues training it on your specific data so it learns a new behavior (style, format, reasoning pattern). It changes how the model behaves, not what it knows. Full explainer →
The application layer
AI Agent A system built around a language model that can take real actions: calling tools, reading results, and adjusting its plan. The difference from a chatbot is action: a chatbot tells you what to do, an agent does it. Full explainer →
Agentic AI AI systems that operate with autonomy: making decisions, taking actions, and coordinating with other agents without human intervention at every step. The buzzword version of “AI agent” that buyers hear in pitches.
MLOps (Machine Learning Operations) The engineering discipline for deploying, monitoring, and maintaining AI models in production. It catches model drift, triggers retraining, and ensures every decision is auditable. Full explainer →
RAG Pipeline The full system that powers RAG: document ingestion, chunking, embedding, vector storage, retrieval, and generation. A well-designed pipeline is the difference between accurate AI and hallucinating AI. Architecture blueprint →
The search and discovery layer
Semantic Search Search that matches by meaning instead of exact keywords. If you search “refund,” it finds “return policy” because the concepts are related. Powered by vector embeddings.
Hybrid Search A search approach that combines keyword search (exact matching) with semantic search (meaning matching). Most production systems use hybrid search for the best of both worlds.
Knowledge Graph A data structure that stores information as nodes (entities) and edges (relationships). Lets AI traverse from one fact to connected facts: “Company A acquired Company B which is regulated by Agency C.”
The optimization layer
GEO (Generative Engine Optimization) The practice of optimizing content so AI engines (ChatGPT, Perplexity, Google AI Overviews) cite you in their answers. SEO gets you on page 1 of Google. GEO gets you into the answer itself. Full explainer →
SEO (Search Engine Optimization) The practice of optimizing content to rank in traditional search engine results. Still important, but increasingly complemented by GEO as users shift to AI engines for answers.
Entity Convergence Ensuring your company’s facts (name, description, location, offerings) are consistent across every platform AI engines learn from: your website, LinkedIn, Wikipedia, Crunchbase, industry directories. Inconsistent facts reduce AI citation probability.
llms.txt A file (like robots.txt) that tells AI models what your site is about and which pages are authoritative. A simple but effective GEO tactic that most sites do not have yet.
The governance layer
Model Drift When a model’s accuracy degrades over time because real-world data has changed from what the model was trained on. Detected by MLOps monitoring and fixed by retraining.
Hallucination When an AI model generates a confident but factually incorrect response. Mitigated by RAG (grounding in real documents) and verification gates.
Human-in-the-Loop (HITL) A design pattern where AI proposes an action and a human approves it before execution. Non-negotiable for sensitive actions (money, customer data, production systems).
Explainability The ability to understand and audit why an AI model made a specific decision. Required by regulators in banking, healthcare, and legal industries. Every flag in our AML system comes with a reason.
The business layer
Chief AI Officer (CAIO) The executive responsible for AI strategy, execution, and governance. Decides where AI creates value, builds the systems, and manages the risks. Full explainer →
AI-First Company A company designed from the ground up to run on AI: knowledge-base operating system, agent workflows, humans at the decisions that matter. Not a company that uses AI, but a company that runs on it.
Knowledge-Base Operating System The central “brain” of an AI-first company: a structured knowledge base that agents read from, write to, and use to make decisions. The infrastructure that turns scattered information into a system.
This glossary is a living document. Updated as new terms enter mainstream use. Have a term you want defined? Let me know.
Want the full picture of how these concepts fit together in a real enterprise? Read How to Implement AI in Enterprise or let’s talk.