GenAIOps applies DevOps principles to generative AI applications across development, deployment, monitoring, and maintenance. It extends familiar infrastructure and delivery practices with prompt engineering, model performance checks, output-quality validation, scalability, security, and orchestration. section 12.0 section 12.3
Use the lifecycle as an iterative delivery loop:
- Ideate by selecting a use case and model, testing prompts, connecting data, and validating the proposed solution. section 12.3
- Build with sample data, inspect results, adjust the solution, and expand to larger datasets only after small tests succeed. section 12.3
- Operationalize by configuring endpoints, monitoring performance, adding alerts, and integrating the solution with an application. section 12.3
- Return to earlier phases when results fail quality checks; validation remains part of every phase. section 12.3
Retail customer-service chatbot
A retail chatbot answers natural-language questions such as “What’s the best backpack for a week-long hike?” by searching product data and returning grounded recommendations. The application combines a searchable information store, an AI model, and familiar monitoring and maintenance tools. Sample data testing helps expose problems early, while grounded generation keeps answers tied to actual product information instead of invented content. section 12.3.1
Table 12.6 Comparison of GenAIOps, LLMOps, and MLOps
|
Framework
|
Focus
|
|---|---|
|
MLOps
|
Extends DevOps to ML models, with reproducible pipelines, automated workflows, and CI/CD integration
|
|
LLMOps
|
Focuses on large language models (LLMs), handling training, deployment, and performance monitoring unique to LLMs
|
|
GenAIOps
|
Includes LLMOps but covers all generative AI, including smaller models and multimodal applications. Handles scalability, security, and orchestration across diverse generative systems.
|
Compare GenAIOps with MLOps and LLMOps: GenAIOps covers generative AI broadly, including language, smaller-model, and multimodal applications. section 12.3
Table 12.7 Workshop approach
|
Aspect
|
Description
|
|---|---|
|
Start with a clear business need.
|
Test your idea with sample data first to check your concept and find potential problems early.
|
|
Use familiar tools.
|
The system uses Cosmos DB for orders and Azure Search for the product catalog, so database professionals can apply their current skills to AI projects.
|
|
Ensure accurate information.
|
The system gets answers from your actual product data rather than making up information. This “grounded generation” approach is critical for business applications.
|
Use the workshop approach to connect business needs, sample-data testing, familiar databases, and grounded generation. section 12.3.1
