AI glossary · Basic · MLOps & LLMOps
MLOps
MLOps is a way of automating and systematically managing the whole process of building machine learning models, deploying them to services, and continuing to improve them while they run. The name means linking model development (ML) and operations (Ops) into one.
About this term
- Level
- Basic · Technical terms you often meet in AI news
- Category
- Data, ops & platforms
- Fields
- MLOps & LLMOps
- Added
- 2026-10-08
How it works
Data is collected and cleaned, the model is trained, its performance is evaluated, and models that pass are deployed to the service. After deployment, the team monitors whether predictions are accurate and whether incoming data has drifted from what the model saw in training. If problems appear, the model is retrained on new data and the process repeats. The core of MLOps is running this loop with automation tools and defined procedures instead of by hand.
The MLOps loop
A model that stays useful
- 1Data preparation
- 2Training
- 3Evaluation
- 4Deployment
- 5Monitoring
- 1. Data preparation Collect and clean
- 2. Training Log experiments
- 3. Evaluation Meets the bar?
- 4. Deployment Put into service
- 5. Monitoring Watch performance and drift
Why it matters
A model that worked well during research can easily lose performance over time once it is running in a real service. This is because when the world changes, the incoming data changes too (data drift). And if there is no record of which data and settings were used to build a model, it is hard to find the cause of a problem or roll back to a previous model when something goes wrong. MLOps provides this reproducibility, traceability and stability.
Good to know
MLOps is not a single tool. It is an umbrella term for a set of components, such as experiment tracking, a model registry, model serving and monitoring, along with the ways of working around them. You do not need everything from the start; teams usually begin with record-keeping and deployment automation and expand step by step. Organizational culture also matters: it only works when data scientists and operations engineers share the same process. When working with LLMs, the extended term LLMOps is sometimes used.
Example
Consider a product recommendation model for an online store. When the seasons change, the products people look for change too, and recommendations start to miss. A team with MLOps in place notices early through monitoring that metrics such as click-through rate are dropping, retrains the model on recent data, evaluates it, and automatically deploys only models that pass the bar. If a new model causes problems, they immediately roll back to a previous version stored in the model registry.