googleapis/python-aiplatform

What it solves

This SDK provides a comprehensive interface for developers to build, scale, govern, and optimize AI agents and generative AI workflows on the Gemini Enterprise Agent Platform (formerly Vertex AI). It simplifies the process of deploying agents, managing prompts, and evaluating model performance within the Google Cloud ecosystem.

How it works

The SDK acts as a client library that connects Python applications to Google Cloud's AI infrastructure. It provides high-level modules for:

• Agent Development: Using the Agent Development Kit (ADK) to define agents with specific tools (functions) and deploying them to a managed runtime.
• Prompt Management: Creating, versioning, and optimizing prompts using zero-shot or data-driven methods.
• Evaluation: Running inference on datasets and calculating metrics like exact match or ROUGE-L to assess model quality.
• Skill Registry: Managing a library of reusable skills that agents can call upon, including versioning and semantic search for skill retrieval.
• Data & Training: Managing tabular, text, image, and video datasets for training custom or AutoML models.

Who it’s for

Cloud developers and ML engineers who are building enterprise-grade AI agents and generative AI applications using Google Cloud's Gemini infrastructure.

Highlights

• Agentic Framework: Integrated ADK for creating agents that can execute Python functions as tools.
• Prompt Optimizer: Tools for both zero-shot and data-driven prompt optimization to improve model responses.
• Managed Skill Registry: A system to create, update, and semantically search for agent skills.
• Evaluation Suite: Built-in capabilities to run inference and evaluate results against reference datasets using multiple metrics.