AI engineering terms, in plain language

Software that uses a language model to plan steps and take actions — calling tools and APIs — to complete a goal with limited human input.

When to use it. Use an agent when a task needs several decisions and actions in sequence, such as qualifying a lead or processing a claim.

A technique that retrieves relevant passages from your own data and passes them to a language model so the answer is grounded in that content.

When to use it. Use RAG when answers must reflect current, private, or citable knowledge.

A neural network trained on large amounts of text that can understand and generate language, code, and structured output.

When to use it. LLMs power chat, summarization, extraction, and reasoning features inside software.

A database that stores embeddings — numeric representations of meaning — and finds the items most similar to a query.

When to use it. Vector databases such as Pinecone, PgVector, and Chroma are the retrieval layer of a RAG system.

Further training a pre-trained model on your own examples so it adopts a style, format, or narrow skill.

When to use it. Choose fine-tuning when prompting and RAG cannot reach the quality you need on a specific task.

AI systems that create new content — text, images, audio, or code — from a prompt.

When to use it. Generative AI is used for copilots, assistants, content pipelines, and multimodal features.

Using language models and agents to complete multi-step business work that fixed rules cannot handle.

When to use it. It fits support, sales operations, document processing, and back-office workflows.

Several specialised AI agents that cooperate on a task, each with its own role and tools, coordinated by an orchestrator.

When to use it. Use multiple agents when one agent cannot reliably hold every role, such as research, drafting, and review.