Cheat sheetGCP-02

Google Cloud Gen AI Stack

GCP GenAI Leader / Google Cloud Gen AI Stack

Map needs to layers: Gemini models run on the Vertex AI platform; Model Garden discovers, Vertex AI Studio prototypes. Default to managed.

Vertex AI
The platformUnified place to build, deploy, secure, and manage ML and generative AI.
Gemini
The modelsGoogle's multimodal foundation-model family, with variants trading speed, cost, and capability.
Model Garden vs Studio
Discover vs designGarden catalogs and deploys models; Studio designs, tests, and tunes prompts.
Managed first
Decision rulePrefer a managed model plus prompting or grounding; tune or customize only when a gap remains.
⚑

For any new idea, prototype in Vertex AI Studio with a Gemini model and grounding before considering tuning or custom training.

Managed pathPolicy-doc assistant: Gemini plus grounding on Vertex AI -- live in weeks.
Custom pathOnly if grounding still misses domain tone and you have quality data to tune on.
Vertex AI = platform; Gemini = model family; Model Garden = catalog.
Vertex AI Studio = design and test prompts with little code.
Default managed + prompt/grounding; tune/customize as a later resort.
vertex-aigeminimodel-gardenvertex-ai-studiomanaged-vs-custom
review in 6d