Discuss your projectLet's talk

Knowledge hub

Practical guides to applied AI for business leaders and teams, written by the people who build it.

Guides7

Glossary

AI glossary.

Plain-language definitions of the terms that come up most in AI projects.

AI agent
An AI system that pursues a goal by planning steps, using tools such as databases or email, and checking its results, rather than only replying to a message.
Large language model (LLM)
A model trained on very large amounts of text to predict and generate language. GPT, Claude, Gemini and Llama are examples.
Prompt
The instructions and context given to a model for a task. Good prompts state the goal, the rules, the format and the information to use.
Token
The unit models read and write, roughly a short word or part of a word. Usage costs and context limits are measured in tokens.
Context window
How much text a model can consider at once, including instructions, retrieved documents and conversation history.
Embedding
A list of numbers that represents the meaning of a piece of text, so that similar meanings sit close together and can be searched.
Vector database
A database built to store embeddings and quickly find the passages closest in meaning to a question.
Retrieval-augmented generation (RAG)
A pattern where relevant passages are retrieved from your own sources and given to the model as evidence before it answers.
Re-ranking
A second pass that reorders retrieved passages by relevance, so the best evidence reaches the model first.
Grounding
Tying a model's output to specific, checkable sources rather than to what it memorised in training.
Hallucination
A fluent, confident output that is not supported by facts or sources. Grounding, evaluation and guardrails reduce it.
Fine-tuning
Further training a model on examples to change its behaviour, such as following a format or style. It changes how a model responds more than what it knows.
Guardrails
Rules and checks around a model that block restricted topics, protect personal data and limit the actions it can take.
Evaluation set
A collection of realistic inputs with expected outcomes, used to measure an AI system before and after every change.
LLM-as-judge
Using one model to grade another's output against a written rubric. Useful at scale, but calibrated against human reviewers.
Human in the loop
A design where people review, approve or correct AI outputs at defined points, especially for high-stakes decisions.
Model Context Protocol (MCP)
An open standard for connecting AI applications to tools and data sources through a common interface.
Prompt injection
An attack where instructions hidden in content, such as an email or web page, try to make an AI system ignore its rules.
Open-weight model
A model whose trained weights are published, so it can be run on your own infrastructure. Llama and Mistral models are examples.
Inference
Running a trained model to produce an output. Inference cost and speed matter most once a system is in production.
Model drift
A drop in model performance over time as real-world data or behaviour changes. Monitoring catches it; retraining fixes it.
MLOps
The practices and tools for versioning, deploying, monitoring and retraining machine learning models reliably.
Answer-engine optimisation
Making sure AI assistants and AI-powered search describe and cite your brand accurately, building on SEO.
Agentic workflow
A business process in which one or more AI agents carry out steps, with tools, checks and human approvals defined in advance.

Have a question these guides don't answer? Ask us.