Vertex AI — Google's ML platform
A unified platform to train, tune, deploy, and call ML models — including Google's Gemini family via the Gemini API. Covers the full ML lifecycle.
is 's ML hub. It bundles everything you need to do machine learning: train a model, save it, deploy it to a URL, call it from your app — all under one roof.
Two main ways to use it. (1) Generative AI: call Google's pre-trained LLMs (Gemini, Imagen for images, Veo for video) via the — no training, just a POST with a prompt. Perfect for chatbots, summarization, RAG, image generation. (2) Custom ML: bring your own data, train a model (you code it, or picks architecture for you), deploy it to a real-time endpoint.
You can stay high-level (AutoML: you upload labeled images, Google trains a classifier, gives you a deploy button) or low-level (custom training on GPUs/TPUs with TensorFlow/PyTorch, full control over the training loop).
Serving: models live in the (versioned), get deployed to Endpoints (autoscaling with CPU/GPU/TPU). Your app calls the endpoint's URL, gets predictions. Endpoints can host multiple model versions for A/B testing.
Comparison: Vertex AI = AWS SageMaker = Azure ML. GCP's differentiator: the tightest Gemini integration and first-class support for Ray (distributed Python), BigQuery ML bridge, and Pipelines (MLOps) built on Kubeflow.
Grounded on https://cloud.google.com/vertex-ai/docs/start/introduction-unified-platform
Next up
Firestore — Serverless NoSQL document database
A real-time NoSQL database with offline sync, live listeners, and granular security rules. Ideal for mobile/web apps that want live updates without building a backend.