Guardian AI

Guardian is a multi-agent AI governance system that ensures ethical, compliant, and secure GenAI usage through prompt filtering, policy enforcement, and audit insights.

Comment

Inspiration

With GenAI rapidly entering enterprise systems, organizations face challenges in ensuring responsible usage, regulatory compliance, and ethical integrity. We were inspired to create Guardian AI, a multi-agent system that actively governs GenAI interactions—offering real-time monitoring, auditing, and advisory—so businesses can innovate confidently and securely.

What it does

Guardian AI is a multi-agent governance platform that:

  • Screens GenAI prompts for risks (Prompt Guard)
  • Audits model outputs for compliance (Output Auditor)
  • Enforces dynamic policy rules (Policy Enforcer)
  • Provides regulatory guidance using RAG (Advisory Agent)
  • Collects user feedback for iterative improvement (Feedback Agent)
  • Allows role-based access and audit dashboards for full traceability

How we built it

  • Frontend: React with AWS Amplify & styled in AWS-themed dark UI
  • Backend: Python (FastAPI) hosted on AWS EC2
  • Agents: Built using LangGraph for multi-agent orchestration
  • LLM: OpenAI GPT-4o Mini for fast agents + optional Claude via Bedrock
  • Storage: S3 for document uploads, DynamoDB for logs and interactions
  • RAG: Embedded compliance documents using LangChain + vector DB
  • Security: Role-based login system (Admin/User)
  • Bonus: Admin panel, feedback chatbot widget, compliance audit logs

Challenges we ran into

  • Implementing context-aware dynamic agent orchestration within time limits
  • Integrating RAG pipelines and LLM tool use with performance in mind
  • Handling document uploads securely and linking to regulatory agents
  • Balancing complexity vs clarity for a judge-friendly interface

Accomplishments that we're proud of

  • Created a complete, AWS-integrated multi-agent governance system
  • Achieved smooth agent coordination using LangGraph
  • Successfully implemented role-based prompt moderation and auditing
  • Delivered a professional UI with dashboards, chat, and document upload
  • Used RAG to provide accurate and up-to-date regulatory responses

What we learned

  • Real-time AI governance is possible with the right orchestration
  • LangGraph is powerful for structured agent workflows
  • Combining RAG + prompt auditing creates strong compliance frameworks
  • UX matters as much as backend intelligence in enterprise adoption

What's next for Guardian AI

  • Add more domain-specific compliance agents (e.g., healthcare, fintech)
  • Expand RAG to use live web data and document version control
  • Introduce anomaly detection for prompt or usage patterns
  • Offer Guardian AI as a plug-and-play compliance layer for enterprise LLMs
  • Integrate CI/CD and Infra-as-Code for scalable enterprise deployment

Built With

Share this project:

Updates