-
-
Agents Workflow
-
Dashboard
-
Report and Analysis
-
Blue team Blocks and log (Autonomous agents)
-
Red team attacks and log (Agents with manual trigger)
Inspiration
Modern cybersecurity systems rely heavily on static rules and manual analysis, which struggle to keep pace with rapidly evolving threats. We were inspired to explore how agentic AI systems, similar to those developed by DeepMind, could enable proactive, adaptive, and continuously improving cyber defense using real-world AI models.
What it does
CyberSentinel is an agent-based cybersecurity platform that evaluates and strengthens system security through coordinated AI agents. Manually triggered Red Team simulations model realistic attack scenarios, while AI Blue Team agents autonomously analyze signals, detect threats, and generate defensive responses. A Gemini-powered orchestrator coordinates agent interactions, evaluates outcomes, and enables continuous improvement of security strategies.
How I built it
The platform was built using Google AI Studio for prompt and logic design and the Gemini API for real agent reasoning. The system architecture includes a central orchestrator, distinct Red and Blue Team agents, structured workflows, and explainable outputs. The frontend was implemented using React with TypeScript, while agent logic and orchestration were handled through modular services integrated with the Gemini API. Version control and collaboration were managed using Git and GitHub.
Challenges I ran into
One of the main challenges was distinguishing true agent behavior from prompt-based simulations and ensuring that each agent had a clear role, logic flow, and explainable output. Managing structured responses from generative models and coordinating multi-agent workflows within limited development time also required careful design trade-offs.
Accomplishments that I'm proud of
- Implemented a real multi-agent architecture using the actual Gemini API
- Designed coordinated Red and Blue Team agent workflows
- Built an explainable, orchestrated decision pipeline rather than a chat-based demo
- Successfully combined autonomous AI behavior with human-in-the-loop control
What I learned
We gained practical experience in designing agentic AI systems, orchestrating multi-agent workflows, and integrating generative models into real applications. We also learned the importance of explainability, safety boundaries, and architectural clarity when building AI-driven security systems.
What's next for CyberSentinel - Gemini Agentic Platform
Future work includes adding persistent agent memory, integrating real-time log ingestion, deploying the platform on Vertex AI, and expanding agent capabilities for automated remediation, compliance analysis, and continuous threat intelligence integration.
Log in or sign up for Devpost to join the conversation.