Skip to main content

AI glossary Operations

What is a pipeline in AI and machine learning?

← All glossary terms
In short

A sequence of processing steps, such as ingestion, preparation, model and post-processing, that turns input data into a result.

Explained

A pipeline is a chain of processing steps in which the output of one step becomes the input of the next: for example, data ingestion, cleaning, feature preparation, model inference and post-processing. In machine learning, pipelines cover both how models are trained and how they are used in production. In generative AI, a pipeline can chain retrieval, prompt construction, model calls, validation and actions. Building AI systems as pipelines makes each step testable, repeatable and replaceable, which is what makes them reliable and scalable.

Types of pipeline

  • Data pipelines move and transform data from source systems into a form that analytics or models can use.
  • Training pipelines prepare datasets, train models, evaluate them and register the versions that pass.
  • Inference pipelines take a live request, prepare the input, call the model and post-process the result.
  • LLM and RAG pipelines retrieve context, build the prompt, call the model, check the output and return or act on it.
  • CI/CD pipelines for ML test and deploy code, data and models together, so that changes reach production in a controlled way.

Why build AI as pipelines

A notebook that works once is not a system. Turning the steps into a pipeline means each one has defined inputs and outputs, can be tested on its own and runs the same way every time.

That makes it possible to retrain models automatically, reproduce a past result, swap one model for another, and see which step failed when something goes wrong. It is also the basis for monitoring, because every step can emit metrics and traces.

Pipeline or agent?

A pipeline follows a fixed sequence that someone designed in advance. An agent decides at run time which steps to take and which tools to use.

Pipelines are more predictable, cheaper and easier to audit, so they suit well-understood, repeatable tasks. Agents suit open-ended tasks where the path cannot be known in advance. Many production systems combine both: a pipeline with a few steps where an agent makes a bounded decision.

Frequently asked questions

What is an ML pipeline?

An ML pipeline automates the steps of building and using a machine learning model: data preparation, training, evaluation, deployment and, in production, inference. Automating them makes results reproducible and lets models be retrained when data changes.

What is a data pipeline?

A data pipeline moves data from one or more sources to a destination, such as a data lake or warehouse, applying transformations on the way. It can run in batches on a schedule or continuously as a stream.

What is a RAG pipeline?

A RAG pipeline is the sequence that turns a question into a grounded answer: retrieve relevant passages, build a prompt with them, call the language model and return the answer with its sources. Ingestion and indexing of documents form a separate pipeline that feeds it.

What is the difference between a pipeline and a workflow?

The terms overlap. Pipeline usually stresses data flowing through a chain of processing steps; workflow is broader and can include human tasks, approvals and branches. In practice, an AI workflow often contains one or more pipelines.

What is MLOps and how does it relate to pipelines?

MLOps applies DevOps practices to machine learning: version control, automated testing, continuous delivery and monitoring. Pipelines are its main building block, since they are what gets automated, tested and deployed.

Decision guides that use this term

Where it shows up at Thinkia

Products

Sources

Updated: 2026-10-21

Want to see how this applies to your organisation?

Talk to an AI Expert