AI glossary

A

Agentic workflow

A business process in which AI components plan or execute some steps, usually alongside fixed steps and human approvals.

AI agent

A system in which a language model decides which actions to take, calls tools to take them and uses the results to choose its next step.

E

Embedding

A numerical vector representing the meaning of text, images or other data, so that similar items have nearby vectors.

Evaluation set

A curated collection of realistic inputs with definitions of correct output, used to measure AI system quality.

F

Fine-tuning

Further training a pre-trained model on task-specific examples to adjust its behaviour, style or format.

G

Grounding

Constraining a model’s output to information in supplied sources, so that claims can be traced and verified.

Guardrails

Runtime checks on the inputs and outputs of an AI system that block, modify or escalate content that breaks defined rules.

H

Hallucination

Output from a generative model that is fluent and plausible but not supported by its inputs or by fact.

Human-in-the-loop

A design in which people review, approve or correct AI outputs or actions at defined points.

Hybrid search

Combining keyword-based and vector-based retrieval, then merging and re-ranking the results.

L

LLM observability

Collecting traces, metrics and quality signals from language model applications to understand behaviour in production.

LLM-as-a-judge

Using a language model to score or compare outputs of an AI system against criteria.

M

Model routing

Sending each request or step to the most suitable model based on difficulty, cost, latency or data sensitivity.

O

Orchestration

Coordinating the steps of an AI-enabled process: model calls, retrieval, tools, rules, approvals, retries and error handling.

P

Prompt injection

An attack in which text supplied to a model, directly or inside retrieved content, attempts to override its instructions.

R

Re-ranking

A second retrieval stage in which a more precise model reorders candidate passages by relevance to the query.

Retrieval-augmented generation (RAG)

A pattern in which relevant passages are retrieved from a body of content and given to a language model to ground its answer.

S

Structured output

Constraining a model to produce data in a defined format, such as JSON matching a schema.

T

Tool calling

A model capability to request structured calls to external functions or APIs, whose results are returned to the model.

V

Vector database

A data store optimised for finding the vectors most similar to a query vector, used for semantic search over embeddings.