My Journey Through Google Cloud GenAI Academy APAC Edition Cohort 3 — 2026

My journey with the Google Cloud GenAI Academy APAC Edition Cohort 3 has been an eye-opening experience, marking my first comprehensive learning opportunity with Google Cloud and AI agents. Prior to joining this academy, I had dabbed in generative AI and AI-driven projects, yet this program offered an unparalleled insight into the capabilities of AI agents in connecting models with tools, data, applications, and cloud infrastructure.

The academy transcended the mere act of watching videos or studying theory, incorporating practical workshops, codelabs, and quizzes, culminating in an Ideathon Prototype Submission where we had to convert our acquired knowledge into a functional project.

The program was meticulously structured into three main tracks: Track 1 centered on constructing and deploying a customer-facing AI agent, Track 2 concentrated on leveraging AI agents to convert business data into strategic decisions, and Track 3 delved into automating daily operations using a productivity agent. Each track featured a blend of theoretical sessions, hands-on codelabs, and quizzes.

In Track 1 — Build and Deploy a Customer-Facing AI Agent, I was introduced to the creation of an AI agent for customer-facing applications. One of the hands-on codelabs centered on deploying a Retrieval-Augmented Generation (RAG) AI agent in Streamlit utilizing Google ADK and Cloud Run. This project centered around a coffee shop scenario, aiming to develop an AI Barista agent capable of answering customer inquiries and offering recommendations based on a menu dataset.

The codelab elucidated critical concepts such as Google Agent Development Kit (ADK) for agent construction, Gemini as the underlying AI model, Retrieval-Augmented Generation (RAG) for grounding responses in external data, a menu dataset encompassing coffee items, tags, and allergens, Streamlit for the chat interface, and Cloud Run for application deployment.

What struck me was the agent's ability to retrieve pertinent information from the menu data, enabling more informed recommendations, including allergen-related details. This was my initial encounter witnessing how an AI agent, RAG, application interface, and cloud deployment could operate in unison.

Track 2 — Turn Business Data into Strategic Decisions focused on harnessing AI agents with business data. The codelab introduced the deployment of AI agents with Gemini and the BigQuery MCP Server on Cloud Run. This segment illuminated how AI agents could interact with structured data rather than solely responding to generic text prompts.

Key technologies included ADK, Gemini, BigQuery, BigQuery MCP Server, Model Context Protocol (MCP), and Cloud Run. A particularly intriguing aspect was the Model Context Protocol (MCP). This standard protocol enables AI applications and agents to interface with external tools and data sources, allowing an AI agent to access and analyze structured business data stored in BigQuery through the BigQuery MCP Server.

This showcased a different facet of AI agents: their potential to serve as intermediaries between users and intricate data systems, transforming raw data into valuable insights and decisions.

The third track, Track 3 — Automate Daily Operations with a Productivity Agent, centered on employing an AI agent for automating operational tasks. A codelab showcased a personal agent in a Cloud Run service, utilizing a coffee shop manager scenario. The agent could analyze business data through a chat interface and execute tasks.

Notably, the use of Cloud Run sandboxes allowed the agent to dynamically generate and execute Python code within a secure, isolated environment to resolve analytical problems. For instance, the coffee shop manager could instruct the agent to analyze Point-of-Sale data alongside a university graduation schedule to pinpoint potential operational inefficiencies, such as the correlation between drink complexity and cashier staffing.

The agent could then formulate recommendations for staffing, inventory management, operational planning, and generate actionable TODOs, albeit requiring explicit authorization to implement changes. This segment reinforced the notion that productivity agents are not confined to answering queries; they can analyze information, execute code, and assist in real-world operational workflows.

Post the completion of the learning tracks, the academy transitioned from knowledge absorption to practical application. The Ideathon Prototype Submission challenged participants to materialize an idea using Google AI and Google Cloud technologies. The submission required a functional prototype accompanied by a social post detailing the solution, technical methodology, and utilization of Google Cloud tools.

The evaluation criteria encompassed authenticity, usability, stability, and security. My submission, Reflectora-AI, was a personal AI journaling application designed to assist users in reflecting on their thoughts and converting them into actionable insights. The application featured different AI modes including Reflect, Summarize, Brainstorm, Action Plan, and Ask Gemini.

Reflect allows users to delve into their thoughts and receive reflective responses, Summarize condenses lengthy journal entries, Brainstorm generates ideas based on user thoughts, Action Plan transforms reflections into practical next steps, and Ask Gemini enables interaction with the AI regarding journal entries.