Glossary

Plain-language definitions of the terms used across our courses and the career & salary map.

Context window

The maximum number of tokens a model can attend to at once.

Embedding

A dense vector representation of text (or other data) whose geometry captures semantic similarity.

Evaluation set

A held-out set of examples used to measure quality objectively.

Feature store

A system for defining, storing and serving model features consistently for training and inference.

Fine-tuning

Continuing training of a pretrained model on task-specific data to adapt its behaviour.

Guardrails

Checks around a model that constrain inputs and outputs for safety, policy or cost.

Hallucination

A confident but unsupported or incorrect model output.

LoRA

Low-Rank Adaptation — a parameter-efficient fine-tuning method that trains small adapter matrices.

MLOps

Practices and tooling for deploying, monitoring and iterating on ML systems in production.

Model serving

Exposing a trained model behind an API with latency, throughput and reliability targets.

Prompt engineering

Designing the text input to an LLM to steer its output reliably.

RAG

Retrieval-Augmented Generation — grounding an LLM’s answers in retrieved documents rather than relying only on its parameters.

Token

The unit an LLM reads and generates — roughly a word piece.

Transformer

The attention-based neural architecture underlying modern language and vision models.

Vector database

A store optimised for nearest-neighbour search over embeddings.