-
-
Aegis Shield Architecture
Inspiration
Healthcare AI holds immense promise — but one HIPAA violation can cost millions and erode public trust. GenAI systems are often “after-the-fact” auditable, but not proactively safe. We were inspired to build Aegis Shield to act as a smart firewall: a real-time guardian that ensures compliance before harm is done.
What it does
Aegis Shield is a multi-agent AI governance layer for GenAI apps used in healthcare. It:
- Intercepts every prompt before it reaches the LLM
- Blocks violations in real time (e.g., PHI exposure)
- Visualizes agent interactions via a live UI
- Generates PDF audit reports for each session
How we built it
- Used AWS Lambda to implement a “Guardian Agent” for real-time prompt filtering
- Integrated Amazon Bedrock (Claude) to process compliant prompts
- Used Streamlit + pyvis to visualize agent interactions
- Stored logs in DynamoDB and S3
- Generated compliance reports with FPDF2 (Python)
All components run serverlessly for easy deployment and low cost.
Challenges we ran into
- Designing a lightweight, real-time policy engine that didn’t block valid queries
- Ensuring compliance logic worked across nuanced healthcare prompts
- Balancing UX clarity with backend complexity in the demo visualizer
- Minimizing false positives without compromising safety
Accomplishments we’re proud of
- Built a complete, multi-agent governance system on AWS in under 2 weeks
- Demonstrated proactive AI safety in a live demo using red/green feedback
- Delivered enterprise-ready features like 1-click audit trail generation
What’s next
- Add user-role access simulation (Doctor vs. Public vs. Researcher)
- Extend compliance logic for GDPR and EU AI Act use cases
- Open-source our agent orchestration SDK for broader adoption
Log in or sign up for Devpost to join the conversation.