AgentSquad

Collection of Agent squads to accomplish an activity

Comment

Inspiration

Hospitals today handle thousands of patient interactions daily — from symptom triage to booking appointments and sending reminders. Most of these are still handled manually, leading to delays, burnout, and missed follow-ups.

I wanted to explore how multi-agent systems powered by AWS Bedrock could automate these workflows, allowing healthcare staff to focus more on care and less on coordination.

My Goal : Build an AI-driven healthcare assistant squad — a team of specialized agents that work together seamlessly to deliver faster, safer, and smarter patient experiences

What it does

AgentSquad is a multi-agent healthcare assistant that automates critical patient interactions: • Triage Agent: Collects symptoms and provides preliminary assessments or advice. • Booking Agent: Finds and schedules appointments with available doctors. • Reminder Agent: Sends follow-ups for medication, lab tests, or vaccinations. • Supervisor Agent: Coordinates tasks among agents, ensuring the right workflow and consistency.

The system stores patient data and conversation history in Amazon DynamoDB, leverages AWS Bedrock Claude models for reasoning, and exposes a Streamlit-based demo interface for interaction.

How we built it

• Architecture: Multi-agent system designed using AWS Bedrock’s Agent framework.
• Backend: AWS Lambda functions act as “tools” for each agent (e.g., fetching patient info, updating bookings).
• Data Layer: Amazon DynamoDB stores patient records, appointments, and chat context.
• Orchestration: A Supervisor Agent dynamically routes user queries to the right sub-agent.
• Frontend: Streamlit web app where users can chat, book appointments, and view reminders.
• Deployment: AWS SAM templates for Lambda + API Gateway; Streamlit hosted in cloud

Challenges we ran into

• Integrating real-time data updates in DynamoDB without race conditions.
• Managing state and memory across multiple agents and sessions.

Accomplishments that we're proud of

• Built a fully functional multi-agent prototype using AWS Bedrock’s latest capabilities.
• Demonstrated autonomous collaboration between agents for triage, booking, and follow-up.
• Achieved seamless integration between LLM reasoning and AWS-native services.
• Created a Streamlit cloud demo that shows real-world healthcare workflows powered by AI.

What we learned

• The power of Agentic architecture in structuring real-world workflows.
• How AWS Bedrock integrates with Lambda and DynamoDB for LLM applications.
• Importance of tool orchestration, memory management, and human handoff design.
• Techniques for prompt optimization, context persistence, and model evaluation.
• That collaboration between multiple AI agents can mirror teamwork in real hospital settings.

What's next for AgentSquad

• Deploy the platform for elder care follow-up and chronic disease management pilots.
• Add Doctor and Nurse Assistants to handle prescriptions and patient instructions.
• Enable voice interaction via Amazon Lex or Transcribe.

Built With

Share this project:

Updates