Multi-Agent Placement and Semantic Query System
About the Customer
A growing pet services company operating dog daycare, boarding, and grooming facilities. Staff made placement decisions manually — evaluating behavioral history, temperament, size compatibility, and social dynamics for each dog entering the facility.
Challenge
Manual placement decisions took 5–10 minutes per dog. Contact tracing for behavioral incidents required 30–60 minutes of searching through paper records. There was no semantic search capability across behavioral audit logs, making pattern recognition across the facility impossible.
Solution
EFS Networks built a multi-agent system using the AWS Strands Agents SDK with three specialized agents:
- Placement Agent — Autonomous reasoning over graph relationships (Neptune Serverless) to match dogs with compatible play groups
- Query Agent — Semantic search across behavioral audit logs using Bedrock Knowledge Bases and OpenSearch Serverless
- Supervisor Agent — Orchestrates the specialized agents, handles multi-turn conversations, and routes requests
Human-in-the-loop confirmations ensure staff approve all placement decisions before execution. The system uses ReAct reasoning patterns for transparent decision-making.
Key AWS Services
AWS Strands Agents SDK, Amazon Bedrock, Amazon Neptune Serverless, Bedrock Knowledge Bases, Amazon OpenSearch Serverless, Lambda, Step Functions, Cognito, WAF.
Let's talk about what you're building.
Our team brings over two decades of experience to every engagement. Tell us about your project and we'll show you what's possible.