Inspiration
- Came directly from real warehouse operations already managed in our WarePro product
- Customers struggle with manual coordination between packing and transport
- Wanted to bring autonomous decision-making into an existing supply chain system
- Built this hackathon project as the first step toward integrating Bedrock Agentcore into WarePro
- The objective was to prove real-world usefulness, not a theoretical demo
What We Learned
- How to design a multi-agent workflow where agents collaborate instead of acting as isolated tools
- How Agentcore memory becomes meaningful when tied to ongoing logistics state and batching logic
- That a serverless event-driven architecture is a natural fit for logistics workloads
- Where 3D packing optimization needs to align with downstream transport booking logic
- How tool design influences clean agent boundaries and specialization
How We Built It
- Three Nova-based Bedrock agents: Orchestrator, Analyser, and Transport
- Orchestrator drives the workflow and decision chain end-to-end
- Analyser performs 3D packing and layout generation
- Transport agent handles capacity selection and booking
- Mock Order and Transport APIs were used so everything could be fully open sourced
- Packing layouts are saved to S3 and surfaced in the control panel
- SNS + Lambda are used for notifications and event triggers
- React + Three.js UI renders the resulting 3D layouts
- Architecture designed so it can plug into WarePro after the hackathon
Challenges
- Designing a fully autonomous flow with zero manual orchestration
- Correct sequencing across three agents using Lambda and SNS
- Preserving a production architecture while still using simulated APIs
- Handling realistic bin-packing constraints (fragility, stacking, weight distribution)
- Making sure the system could scale without installing or managing any infrastructure
- Building everything within a short timeline while keeping it extensible
Log in or sign up for Devpost to join the conversation.