Koala Fleet

• 

  Screen Koala Fleet

• 

  Result screen 1

• 

  Result screen 2

• 

  Security

• 

  Audit Logs

Inspiration

In traditional enterprise environments, cross-department onboarding workflows between HR, Finance, and IT are highly fragmented, slow, and prone to security and PII leaks. We were inspired to build Koala Fleet to demonstrate how an autonomous, multi-agent AI system can orchestrate end-to-end corporate workflows instantly from a single request while enforcing strict enterprise security guardrails.

What it does

Koala Fleet takes a natural language request (such as onboarding a new team member) and orchestrates specialized agents in parallel:

• "HR Agent:" Generates tailored job descriptions and role postings.
• "Finance Agent:" Verifies budget limits and approves expenditure allocations.
• "IT Agent:" Provisions necessary hardware (e.g., MacBook Pro setup) and initiates cloud system permissions.
• "Security Guardrails:" Integrates Model Armor to intercept PII leakage and block prompt injection attacks in real time, logged cryptographically under an Audit Logger. ## How we built it
• "Framework & UI:" Built with Python and Streamlit for an interactive multi-agent control panel.
• "LLM Engine:" Powered by Google's Gemini 2.5 Flash model for agent reasoning and task execution.
• "Agent Orchestration:" Deployed using Google Cloud ADK (Agent Development Kit) architecture.
• "Security Layer:" Guarded by Model Armor, Agent Gateway controls, and real-time Audit Logging.
• "Hosting:" Run and tested within Google Cloud Shell environment. ## Challenges we ran into -Coordinating state management and parallel execution outputs across multiple autonomous agents. -Ensuring real-time PII redaction without degrading Gemini's response speed or agent accuracy.

Accomplishments that we're proud of

-Built a working multi-agent enterprise platform within a single unified control dashboard.

• Successfully implemented active PII redaction and adversarial prompt protection using Model Armor principles. ## What we learned
• Deepened our understanding of orchestrating agentic workflows using Google Cloud ADK.
• Learned best practices for securing multi-agent networks against data leakage and injection risks. ## What's next for Koala Fleet
• Expanding integrations with enterprise SaaS systems like Slack, Jira, and Workday.
• Adding automated policy enforcement and customized multi-tenant RBAC permissions.

Built With

• adk
• ai-agents
• gemini
• google-cloud-shell
• model-armor
• python
• streamlit