CloudSentinel AI

• 

  Excited to share my latest project: CloudSentinel AI — An Autonomous Incident Self-Healing Swarm built for cloud security!

Inspiration

Cloud security incidents and server downtimes cost businesses billions annually. When critical vulnerabilities or configuration drifts occur in cloud infrastructure, traditional monitoring systems trigger alerts, but human engineers still take hours to investigate logs, write patches, test scripts, and deploy fixes. We built CloudSentinel AI to automate this entire lifecycle using autonomous multi-agent systems powered by Google Cloud & Gemini.

What it does

CloudSentinel AI is an autonomous, self-healing cloud security and infrastructure remediation agent swarm: Monitor Agent: Scans Google Cloud logs and telemetry in real-time for security breaches, unauthorized access, or misconfigurations. Analyst Agent: Uses Gemini 1.5 Pro to analyze root causes, consult security documentation, and generate executable repair scripts. Verifier & Execution Agent: Automatically dry-runs fixes inside an isolated Google Cloud Run environment, validates security metrics, applies approved patches, and generates detailed incident summaries.

How we built it

Brain & Orchestration: Google Vertex AI + Gemini 1.5 Pro with native Function Calling for agent reasoning and tool execution. Infrastructure: Google Cloud Run for serverless execution and Eventarc for cloud event triggers. Dashboard: Built with Python & Streamlit, giving engineers real-time visual tracking of multi-agent reasoning and logs.

Challenges we ran into

Agent State Management: Preventing infinite retry loops during automated script testing and ensuring strict execution bounds. Safety & Sandboxing: Ensuring autonomous code generation is tested safely in isolated containers before making changes to live cloud infrastructure.

Accomplishments that we're proud of

Building a zero-human-intervention cloud remediation workflow that reduces incident response time from hours to under 60 seconds. Crafting an intuitive UI that visualizes multi-agent reasoning, tool calls, and patch verification in real-time.

What we learned

Gemini Function Calling: Deepened our understanding of leveraging Gemini 1.5 Pro's native function calling to build deterministic agent execution pipelines. Autonomous Error Remediation: Learned how to safely sandbox multi-agent execution loops inside Google Cloud Run without risk to production resources. Agentic State Management: Discovered effective strategies for maintaining persistent context across multiple collaborating AI agents.

What's next for CloudSentinel AI

Expand support for multi-cloud environments (AWS & Azure).Add interactive Human-in-the-Loop approval toggles for high-impact production changes.

Built With

• cloud-run
• gemini-api
• google-cloud
• python
• streamlit
• vertex-ai