AI Dictionary of Terms

MLOps / LLMOps

A set of practices, tools, and cultural principles that combine Machine Learning, DevOps, and Data Engineering to reliably deploy, monitor, version, and maintain ML models in production — with LLMOps being the specialized subset focused on the unique challenges of Large Language Models (prompts, evaluations, RAG pipelines, and context management).

The Simple Version

Imagine you’ve built an amazing race car in your garage (the ML model). It’s fast, it’s beautiful, and it wins every race in testing. But to actually use it in a real racing league, you need a whole support system: a pit crew, fuel logistics, spare parts inventory, telemetry monitoring, and a process for upgrading it between races.

That support system is MLOps. It’s everything that happens after the model is built to keep it running reliably in the real world.

LLMOps is the same idea, but specifically for Large Language Models. LLMs have unique needs — you have to version prompts (not just code), evaluate subjective outputs (not just accuracy), manage RAG knowledge bases, and track token costs. LLMOps is the specialized discipline that handles these new challenges.

Detailed Explanation

MLOps emerged around 2018 as organizations realized that deploying ML models was fundamentally different from deploying traditional software. Models degrade over time, depend on data that changes, and require continuous experimentation.

The MLOps Lifecycle (Cradle-to-Grave):

1. Experimentation & Development:

2. Model Registry & Versioning:

3. CI/CD for ML:

4. Feature Management:

5. Monitoring & Observability:

6. Governance & Compliance:

LLMOps: The LLM-Specific Layer: LLMs introduce new challenges that traditional MLOps doesn’t address:

Challenge Traditional MLOps LLMOps
Versioning Model weights Model + prompts + RAG data + context templates
Evaluation Accuracy, F1, RMSE LLM-as-judge, human eval, task-specific benchmarks
Data Static datasets Dynamic knowledge bases, vector stores
Cost Compute time Token usage (input + output)
Safety Model robustness Prompt injection defense, guardrails
Debugging Error logs Trace chains of thought, tool calls, retrievals

Popular MLOps/LLMOps Tools:

MLOps Platforms:

LLMOps Platforms:

Key Characteristics

Business Context

MLOps/LLMOps is the difference between AI demos and AI value:

Why It Matters:

Enterprise Maturity Levels:

Level Description Characteristics
Level 0 Manual Notebooks, ad-hoc deployments, no monitoring
Level 1 Basic CI/CD Automated training pipelines, basic model registry
Level 2 Advanced Full CI/CD, monitoring, drift detection, feature stores
Level 3 Optimized Continuous training, automated retraining, A/B testing

Cost Justification:

Organizational Impact:

Real-World Analogy

A professional restaurant kitchen vs. home cooking. At home, you can cook a great meal with minimal systems. But to run a restaurant that serves 500 customers nightly, consistently, safely, and profitably — you need recipes (model registry), inventory management (feature store), quality control (monitoring), staff training (CI/CD), and health inspections (governance). MLOps is the professional kitchen system for AI.

Code Example

# Simple MLOps workflow with MLflow
import mlflow
import mlflow.sklearn
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# Set experiment
mlflow.set_experiment("customer-churn-model")

# Prepare data
X, y = make_classification(n_samples=10000, n_features=20, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

# Train with experiment tracking
with mlflow.start_run(run_name="rf_v1_baseline"):
    # Log parameters
    n_estimators = 100
    max_depth = 10
    mlflow.log_param("n_estimators", n_estimators)
    mlflow.log_param("max_depth", max_depth)
    
    # Train model
    model = RandomForestClassifier(
        n_estimators=n_estimators,
        max_depth=max_depth,
        random_state=42
    )
    model.fit(X_train, y_train)
    
    # Evaluate
    y_pred = model.predict(X_test)
    accuracy = accuracy_score(y_test, y_pred)
    
    # Log metrics
    mlflow.log_metric("accuracy", accuracy)
    mlflow.log_metric("test_samples", len(X_test))
    
    # Log the model itself (with signature for serving)
    signature = mlflow.models.infer_signature(X_train, model.predict(X_train))
    mlflow.sklearn.log_model(model, "model", signature=signature)
    
    # Set tags for governance
    mlflow.set_tag("owner", "data-science-team")
    mlflow.set_tag("stage", "staging")
    mlflow.set_tag("data_version", "v2026.08.13")
    
    print(f"Model accuracy: {accuracy:.4f}")
    print(f"Run ID: {mlflow.active_run().info.run_id}")

# The model is now in the registry, versioned, tracked, and ready for deployment
# A CI/CD pipeline can automatically promote it to production after validation

Common Misconceptions

Sources & Further Reading