AI glossary — LLM, RAG & AI agentsAI, explained simply.
20 terms that come up in every AI project — short, correct and without marketing.
- Large language model (LLM)
- A neural network trained on vast amounts of text that can understand and generate language. Examples include GPT models or open models that can run locally.
- Retrieval-augmented generation (RAG)
- A method where relevant passages from your own data are retrieved and handed to the language model before each answer, so the AI answers from your documents and can cite sources.
- AI agent
- An AI application that does not just answer but plans steps and uses tools such as databases, email or APIs on its own to complete a task.
- Embedding
- A list of numbers (a vector) representing the meaning of a text. Similar content sits close together in vector space — the basis of semantic search.
- Vector database
- A database that stores embeddings and finds the most semantically similar entries very quickly. A core building block of RAG systems.
- Function calling
- The ability of a language model to produce structured calls to defined functions — such as “find customer” or “create appointment”. This is what makes agents possible.
- Prompt
- The instruction or input given to a language model. Good prompts include role, context, task and desired format.
- Hallucination
- A plausible-sounding but false statement by a language model. RAG, mandatory citations and automated checks reduce the risk considerably.
- Local LLM / on-premise AI
- A language model running on your own hardware instead of someone else’s cloud. Benefit: full data control and predictable cost.
- Quantisation
- A technique that reduces a model’s numeric precision (e.g. from 16 to 4 bits). The model becomes smaller and faster, usually with little loss in quality.
- Inference
- Running a trained model to produce an answer — as opposed to training it.
- Token
- The unit in which language models process text; roughly a word fragment. Cost and speed are often measured in tokens.
- Context window
- The maximum amount of text (in tokens) a model can take into account in a single request.
- Fine-tuning
- Further training a model on your own examples to improve style or specialist knowledge. For factual knowledge, RAG is usually the better choice.
- Computer vision
- The field of AI that analyses images and video — for quality control, counting or text recognition (OCR). Tools: OpenCV, TensorFlow.js.
- Genetic algorithm
- An optimisation method modelled on evolution: solutions are scored, combined and mutated until a very good one is found.
- Data mining
- Finding patterns, relationships and segments in large datasets, e.g. for forecasts or customer analysis.
- WebSocket
- A protocol for a persistent, two-way connection between browser/app and server — the basis for chats, live dashboards and tracking.
- VoIP / Asterisk
- Voice over IP is telephony over the internet. Asterisk is open-source PBX software for IVR, queues and AI assistants.
- GDPR-compliant AI
- AI use where legal basis, data minimisation, processing agreements, storage location and deletion policy are settled. Local models simplify this considerably.
Ready when you are
Let’s talk.
Describe your project in a few sentences — you’ll get an honest assessment of feasibility, effort and the way forward.
- info@quell-code.com
- Phone
- Call-back on request
- Response time
- usually 2 working days
- Languages
- Deutsch · English · Türkçe