What to Build

Build and deploy a next-generation, autonomous AI Agent leveraging Gemini 3.5 Flash that operates beyond standard chat loops. The system can run asynchronously in the background, handle the heavy lifting of complex workflows, or dynamically manipulate data pipelines and representations.

Projects must be built within one of these three categories: 

1. Taskmaster: Build a complete workflow, not just a chatbot. Don't just make an agent that writes text. Make one that takes action. Find a messy, multi-step chore in your job, classes, or personal life. Build an agent that handles the details, sends the right info to the right places, and proves it can do the heavy lifting for you.

2. Collaborative Partner: Build an agent that leads the way and takes notes. It should ask clarifying questions, guide the user step-by-step, and have a clear way to capture feedback, so it constantly adapts to the user's unique way of thinking.

3. Fortified Enterprise Fleet: Build a scalable network of institutional agents that hook into official enterprise infrastructure. Teams must demonstrate how agents are cataloged for cross-department use, how they safely maintain context across weeks of asynchronous operations, and how they interact with production data without violating enterprise compliance, data sovereignty, or security policies.

   • Discovery & Lifecycle: Agent Registry (the central repository for publishing, versioning, and discovering enterprise-approved agents).

   • Core Execution & State: Agent Runtime (for long-running, asynchronous background execution) and Memory Bank (for persistent, secure cross-session context over extended timelines).

   • Security & Governance: Agent Identity (For zero-trust access control), Agent Gateway (for unified routing and policy enforcement), and Model Armor (inline guardrails to block prompt injection, tool poisoning, and PII leaks).

   • Telemetry: Agent Observability (OpenTelemetry-compliant audit logs and end-to-end reasoning chain traces).

• Recommended Tech to use (Gemini Enterprise Agent Platform): 

Every project, in every track, must use:

1. Gemini 3.5 or newer accessed through Gemini API or Vertex AI

2. At least one Google Agent Framework: Google ADK, GenAI SDK, Antigravity SDK or GenKit 

3. At least one Google Cloud infrastructure service (such as Cloud Run, Cloud SQL, Firestore, GKE, Pub/Sub).

Note on cost & deployment: Your app does not need to be publicly accessible or live at the exact moment of submission or judging (so you don't rack up unnecessary costs). You just need to provide clear proof that it was built and deployed on Google Cloud — for example, shown in your demo video and code repository. See Resources for tips on keeping your costs near zero.

What to Submit 

• Category 

• URL to the hosted Project (if available) for judging and testing, such as web UI, Chrome Extension, mobile app, etc. A hosted project is highly encouraged.

• Text description 

• Features and functionality

• Technologies used

• Other data sources used

• Findings and learnings 

• URL to your public or private code repository to show how your project was built. If your repo is private, share it with testing@devpost.com and cloudhackathons@google.com 

• Spin-up Instructions: A step-by-step guide in your README.md explaining how to set up and run the project locally or deploy it to the cloud. Even if the judges do not run it, these instructions prove the project is reproducible.

• Architecture Diagram with a clear visual representation of your system (e.g., how Gemini connects to your backend, database, and frontend).

• ~ 4-min Demo video 

• Short overview of the problem your Project is solving

• Value proposition 

• Demo of the app in action

• Must demonstrate the backend is running on Google Cloud (ie: Google Cloud Console, Cloud Run dashboard, Vertex AI logs, URL of .run, etc)

Questions? Start with the FAQs for the quick answers on eligibility, tracks, credits, and submissions and see the Official Rules for anything binding.