Logistics disruptions rarely stay isolated. A delayed shipment can quickly become an SLA breach, customer issue, or operational cost — but recovering from it often means manually gathering information, assessing options, coordinating with people, and tracking what happened.
We built SCÉANCE — “Always watching. Ready to recover.” — to explore what happens when an AI agent takes on that recovery workflow.
SCÉANCE monitors simulated shipments and, when a disruption occurs, investigates the operational context, assesses business consequences, evaluates recovery options, and prepares a recommendation. A human operator remains in control: recovery actions require approval before execution, and the system verifies the resulting state afterward.
The project combines Strands Agents and Claude for agentic investigation and orchestration with deterministic recovery logic for evaluating and executing operational actions. Amazon Bedrock AgentCore, AWS Lambda, EventBridge, and DynamoDB provide the event-driven infrastructure and persistent state needed to support the workflow.
One of our biggest challenges was making the agentic workflow reliable beyond a single execution. We had to design persistent run-scoped state, approval handling, audit events, and session restoration so that an interrupted recovery could continue safely rather than starting over.
Building SCÉANCE taught us that an effective operational AI system is not just about giving an LLM more tools. It is about defining where AI adds flexibility, where deterministic logic provides control, and where humans must remain in the loop.
The result is a working end-to-end recovery workflow:
Disruption → Investigation → Evaluation → Recommendation → Human Approval → Execution → Verification
Built With
- amazon-web-services
- python
- strands
Log in or sign up for Devpost to join the conversation.