statmike/vertex-ai-mlops

What it solves

This project provides a comprehensive library of practical, interactive workflows for implementing machine learning operations (MLOps) and generative AI on Google Cloud Platform (GCP). It bridges the gap between theoretical documentation and production-hardened code by offering adaptable starting points for training, serving, and managing AI models and agents.

How it works

The repository consists of over 470 interactive notebooks organized into specialized tracks:

• MLOps: Covers the full lifecycle from training to production, including serving (vLLM, Triton), feature stores (managed and Bigtable-based), KFP pipelines, and model monitoring.
• Data + AI: Integrates AI functions directly into data services like BigQuery, Dataflow, and Dataproc to bring inference to where the data resides.
• Applied GenAI: Implements advanced RAG pipelines, multimodal prompting, vector search across 11 different GCP databases, and grounding verification.
• Applied Forecasting: Provides a learning path for time series forecasting using everything from BQML to foundation models like TimesFM.
• Framework Workflows: Demonstrates end-to-end patterns for PyTorch, Keras, CatBoost, and R, sharing a common Model Registry and prediction endpoint.
• Applied ML: Features prototypes for AI agents (using ADK and MCP), document processing, and black-box optimization via Vertex AI Vizier.

Who it’s for

Data scientists, ML engineers, and cloud architects who are building AI applications on Google Cloud and need concrete, hands-on examples of how to orchestrate various GCP services for real-world AI workflows.

Highlights

• Massive Scale: Over 470 interactive notebooks covering a vast array of GCP AI services.
• Deep Integration: Shows how to combine Vertex AI with BigQuery, Spanner, AlloyDB, and GKE.
• Agentic Focus: Includes modern workflows for AI agents using the Agent Engine and ADK.
• RAG Specialization: Detailed guides on chunking, ranking, and vector search across multiple database backends.