← Insights
Reference
AI glossary
Working definitions for AI systems, integration and automation. Each term explains its use and a practical limit.
- LLM
- Large language model. It predicts text from context and can help draft, classify or extract information. It does not automatically know your organisation’s current data.
- RAG
- Retrieval-augmented generation. An application searches sources and gives passages to a model as context. Search quality, permissions and citations still need evaluation.
- Embedding
- A numerical representation of text, images or other objects used to compare proximity. Semantic similarity does not prove factual accuracy.
- AI agent
- A system in which a model can choose steps or tools to pursue a goal. Available tools, validation and permissions determine what it can actually do.
- Token
- A unit into which text is split for model processing. A token is not necessarily a word; token counts affect available context and usage cost.
- Inference
- Running a prepared model to produce a prediction or answer, distinct from training it. Both may require infrastructure and evaluation.
- OCR
- Optical character recognition converts characters in an image or scan into text. Fields, reading order and extracted values still need validation.
- API
- An interface through which programs exchange requests and responses under an agreed contract. In automation it defines permitted operations.
- Webhook
- A notification sent from one system to another when an event occurs. Check authenticity and handle duplicates, retries and arrival order.
- Vector database
- A structure for storing and searching numerical representations by similarity. It can support RAG but does not replace source documents or access control.
- Fine-tuning
- Additional training of a model’s parameters for a task. It can change behaviour but does not itself solve frequent document updates or permissions.
- Multimodal
- A system’s ability to process more than one input or output type, such as text, images or audio. Assess quality by modality and task.
- RPA
- Robotic process automation operates other applications through their interfaces. It can help where no API exists, but may depend on screens and flows that change.
- MCP
- Model Context Protocol connects AI applications to tools and data sources through a common interface. The protocol does not grant permissions; applications still authenticate and authorise access.
Want to see these ideas in a process?
Our guides show how sources, models and applications connect in a system that can be evaluated.
All Insights ↗