AxonBrief is a multi-agent digital employee built with Google ADK and Gemini.
It safely researches healthcare AI topics by combining curated knowledge, web search, and structured reasoning tools.
The system uses a Manager → Researcher → Summarizer swarm architecture to generate concise executive briefs while enforcing guardrails that prevent personal medical advice.
AxonBrief demonstrates how agent-based AI systems can safely assist knowledge exploration in healthcare domains.
Healthcare professionals, researchers, and students often face information overload when exploring complex topics such as AI in healthcare.
AxonBrief demonstrates how a multi-agent digital employee can assist by:
- structuring research automatically
- combining curated knowledge with web search
- summarizing complex topics into concise executive briefs
- enforcing safety guardrails to prevent personal medical advice
- providing explainable agent reasoning for transparency
The project highlights how agent-based AI systems can support safe, structured, and responsible knowledge exploration in healthcare domains.
-
Multi-Agent Architecture
Manager, Researcher, and Summarizer agents collaborate to complete tasks. -
Guardrail Safety System
A safety tool blocks requests that attempt to obtain personal medical advice. -
Structured Research Workflow
The researcher agent gathers information from multiple sources. -
Curated Knowledge Base
Internal curated healthcare notes support reliable baseline knowledge. -
Web Search Integration
Public information is retrieved to provide up-to-date context. -
Explainable Agent Outputs
Each agent’s reasoning and actions are visible in the final response. -
Demo-Safe Mode
The system automatically falls back to a demo-safe response if Gemini API quotas are exceeded.
The agent swarm performs structured research using:
- internal curated knowledge
- web search
- research framework tools
The Manager Agent coordinates the Researcher and Summarizer agents to produce the final executive brief.
If a user asks for personal medical advice, the guardrail tool blocks the request and returns a safe response instead.
This ensures the system provides educational information only, not clinical recommendations.
flowchart TD
U[User] --> UI[FastAPI Web UI]
UI --> M[Manager Agent]
M --> G[Guardrail Check]
G -->|Allowed| R[Researcher Agent]
G -->|Restricted| SAFE[Safe Response]
R --> F[Research Framework Tool]
R --> L[Local Knowledge Base]
R --> W[Web Search Tool]
F --> R
L --> R
W --> R
R --> S[Summarizer Agent]
S --> B[Executive Research Brief]
B --> UI
SAFE --> UI