Back to AI Glossary
Developer ExperienceMLOps

MLOps

The set of practices and tools for deploying, monitoring, and maintaining machine learning models in production — essentially DevOps principles applied to the ML lifecycle.

Core Components

MLOps covers the full lifecycle from trained model to production system and back. Model versioning and registry tracks which model version is deployed where, with full lineage back to the training data and hyperparameters that produced it — you need to reproduce any model on demand. Deployment and serving handles the infrastructure for getting predictions out of models at scale, whether that's batch inference, real-time API endpoints, or edge deployment. Monitoring and drift detection watches for the moment production data diverges from training data — models degrade silently when the world changes underneath them. Retraining pipelines automate the response: detect drift, trigger retraining on fresh data, validate the new model, and roll it out with minimal human intervention.

Key Tools

The MLOps tooling ecosystem has matured significantly. MLflow is the open-source Swiss Army knife — experiment tracking, model registry, and deployment in one package. Kubeflow handles orchestration for teams already invested in Kubernetes. Weights & Biases has become the default for experiment tracking and visualisation, particularly popular in research-to-production workflows. Cloud providers offer integrated platforms: AWS SageMaker, Google Vertex AI, and Azure ML bundle training, serving, and monitoring into managed services. The trade-off is always flexibility versus convenience — managed platforms are faster to start but harder to customise as requirements become complex.

Evolution to LLMOps

The rise of large language models has shifted the MLOps conversation. Traditional MLOps assumes you're training models from scratch — LLMOps assumes you're mostly calling APIs or fine-tuning foundation models. The new operational concerns are different: prompt management replaces feature engineering, evaluation frameworks replace traditional ML metrics (how do you measure "good" for open-ended text?), guardrails handle safety at inference time, and cost monitoring becomes critical when every API call has a per-token price. Tools like Langfuse, LangSmith, and Braintrust have emerged specifically for this LLM-era operational stack, focusing on tracing, evaluation, and prompt iteration rather than traditional model training pipelines.

Related Terms

Large Language Model(LLM)— A neural network trained on massive text corpora that can understand and generate human language, typically with billions of parameters.Observability— The practice of monitoring, tracing, and evaluating AI system behavior in production — including LLM calls, latency, costs, retrieval quality, and output correctness.Fine-tuning— The process of further training a pre-trained model on a smaller, task-specific dataset to improve its performance on that particular task or domain.

On the AI Radar

Langfuse— Open-source LLM observability and evaluation platform. Provides tracing, prompt management, evaluation pipelines, and cost tracking for AI applications in production.LangSmith— LangChain's proprietary observability and evaluation platform for LLM applications and agents. Provides tracing, debugging, evaluation (human, heuristic, LLM-as-judge), cost tracking, and deployment across managed cloud, BYOC, and self-hosted options.Braintrust— AI observability and evaluation platform focused on helping teams understand not just what happened, but how to fix it. Combines tracing, evaluation, and deployment gating for production AI.DeepEval— Open-source LLM evaluation framework with 14+ research-backed metrics. Takes a testing-first approach — "pytest for LLMs" — integrating directly into CI/CD pipelines.
Back to AI Glossary