Multi-Agent Warehouse Optimization
This challenge tasks developers with creating a multi-agent system for autonomous warehouse monitoring and optimization, inspired by 'curious AI' for drones. Using Mastra AI, you will design a team of agents that simulate real-time observation, inventory management, and logistical pathfinding. The system will integrate real-time (simulated) data from 'curious' camera/drone agents, allowing for dynamic adjustments to inventory placement and pick-up routes. Mastra AI's built-in memory and tool-use capabilities will be crucial for agents to maintain a consistent understanding of the warehouse state and interact with simulated external systems (like inventory databases or drone control APIs). Claude Sonnet 4 will power the agents' real-time decision-making and provide a human-readable interface through Coplay AI for supervisors to monitor and, if necessary, intervene in operations. BentoML Cloud will be used to deploy and serve simulated 'curious AI' inference modules, demonstrating scalable edge intelligence.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a multi-agent system using Mastra AI for autonomous warehouse monitoring and optimization, integrating real-time insights from 'curious AI' models served by BentoML Cloud, with Claude Sonnet 4 for decision-making and Coplay AI for human interaction.
How work is evaluated
The evaluation will assess the Mastra AI system's ability to simulate effective warehouse monitoring, identify and resolve inventory discrepancies, and optimize logistical paths. Key metrics include response time to simulated events, accuracy of inventory tracking, and efficiency of pathfinding decisions, as well as the robustness of the human-in-the-loop interface.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
Agent State Consistency
Mastra AI agents maintain consistent internal state and memory across interactions.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Tool Use Efficacy
Agents successfully use simulated tools to interact with the warehouse environment.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Resolution Time (seconds)
Average time taken by agents to resolve simulated discrepancies. • target: 90 • range: 30-300
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Path Optimization Ratio
Efficiency of optimized paths compared to baseline non-optimized paths (lower is better for cost/time). • target: 0.75 • range: 0.5-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Master Mastra AI for building stateful, multi-agent systems, leveraging its built-in memory management and asynchronous agent communication patterns.
Design and implement tool-use capabilities within Mastra AI agents to interact with simulated warehouse management systems (e.g., inventory databases, drone control APIs).
Deploy and serve lightweight 'curious AI' inference models (e.g., object detection for inventory) using BentoML Cloud, demonstrating scalable edge intelligence for real-time data capture.
Integrate Claude Sonnet 4 as the reasoning engine for Mastra AI agents, enabling sophisticated decision-making for inventory optimization and dynamic pathfinding.
Build a real-time human-in-the-loop interface using Coplay AI, allowing human supervisors to monitor agent activities, receive alerts, and provide directives to the Mastra AI agents.
Implement dynamic task allocation and coordination strategies among Mastra AI agents to respond to real-time changes in warehouse conditions (e.g., new shipments, misplaced items).
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the Mastra AI system's ability to simulate effective warehouse monitoring, identify and resolve inventory discrepancies, and optimize logistical paths. Key metrics include response time to s...
Challenge input
JSON object with 'discrepancy': {'item_id': str, 'expected_location': str, 'actual_status': 'missing'}
Agent execution
No agent tool is listed for this challenge.
Evaluated output
JSON object with 'resolution_steps': [str], 'final_location': str, 'time_to_resolve_seconds': int
- Mastra AI agents maintain consistent internal state and memory across interactions.
- Agents successfully use simulated tools to interact with the warehouse environment.
- Average time taken by agents to resolve simulated discrepancies. • target: 90 • range: 30-300
- Resolution Time (Seconds) target: 90
- 2 public reference cases
- TypeScript execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Agent State Consistency · Weight 1
- Tool Use Efficacy · Weight 1
- Resolution Time (seconds) · Weight 1
- Path Optimization Ratio · Weight 1
Metrics
- Resolution Time (seconds) · Target 90
- Path Optimization Ratio · Target 0.75
Reference cases
- 2 public reference cases
Workflow tasks
- Task 1
Input: JSON object with 'discrepancy': {'item_id': str, 'expected_location': str, 'actual_status': 'missing'}
Output: JSON object with 'resolution_steps': [str], 'final_location': str, 'time_to_resolve_seconds': int
Execution harness: TypeScript
- Task 2
Input: JSON object with 'current_path': [str], 'new_request': {'item_id': str, 'location': str, 'priority': 'high'}, 'obstacle_location': str
Output: JSON object with 'optimized_path': [str], 'path_cost_units': int, 'rationale': str
Execution harness: TypeScript
Recipe freeze marker
No recipe freeze time is recorded.