Intelligent Edge Operations
Design and implement a cutting-edge multi-agent system to automate and optimize network operations (NOC) within a simulated retail or factory environment. This challenge focuses on building intelligent agents that can monitor real-time sensor data, interact with enterprise systems (e.g., inventory, CRM, facility management), and trigger automated actions at the edge. The solution must demonstrate robust tool integration, adaptive reasoning capabilities for resource-constrained environments, and graph-based workflow orchestration. Participants will leverage the Model Context Protocol (MCP) for seamless and secure tool invocation, allowing agents to interact with diverse simulated edge devices and enterprise APIs. Agentkit will be essential for orchestrating complex, stateful agent workflows, enabling the system to handle unexpected events, perform predictive maintenance, and optimize operational efficiency in real-time. This challenge emphasizes practical, real-world application of advanced agentic AI in an industrial or commercial setting.
What you are building
The core problem, expected build, and operating context for this challenge.
Build a graph-based multi-agent system using Agentkit and MCP for real-time edge operations in a retail or factory setting, leveraging GPT 5 and Gemini 2.5 Pro.
How work is evaluated
Participants will submit a containerized (Docker) agent system designed to handle a series of simulated real-world edge scenarios. The evaluation will focus on the system's ability to accurately interpret events, make correct decisions, utilize tools effectively, and manage resources efficiently.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Orchestrate hybrid reasoning across local and cloud LLMs for optimal performance under edge constraints.
Develop strategies for adaptive thinking budgets to manage compute and response times at the edge.
Implement MCP-enabled agents for secure and dynamic tool integration with simulated enterprise and IoT systems.
Design a robust, graph-based agent workflow for real-time edge operations.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
Participants will submit a containerized (Docker) agent system designed to handle a series of simulated real-world edge scenarios. The evaluation will focus on the system's ability to accurately interpret events, make...
Challenge input
{ 'scenario_id': 'string', 'sensor_data': {'temperature': float, 'pressure': float, 'vibration': float}, 'image_feed': 'base64_encoded_image', 'tim...
Agent execution
No agent tool is listed for this challenge.
Evaluated output
{ 'action_taken': 'string', 'tool_invocations': [{'tool_name': 'string', 'parameters': {}}], 'explanation': 'string', 'resource_usage': {'cpu_cycle...
- Agent takes the appropriate primary action for the scenario.
- All invoked tools are valid and parameters are correctly formatted.
- Evidence of LangGraph's state management (e.g., via logs or specific output fields indicating multi-step re...
- DecisionAccuracy target: 0.95
- Docker 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 takes the appropriate primary action for the scenario.
- All invoked tools are valid and parameters are correctly formatted.
- Evidence of LangGraph's state management (e.g., via logs or specific output fields indicating multi-step reasoning) is present.
Metrics
- DecisionAccuracy · Target 0.95
- ToolIntegrationEffectiveness · Target 0.9
- ResourceEfficiency (LLM Tokens) · Target 3000
- ResponseLatency (Simulated CPU Cycles) · Target 1000000
Workflow tasks
- Task 1
Input: { 'scenario_id': 'string', 'sensor_data': {'temperature': float, 'pressure': float, 'vibration': float}, 'image_feed': 'base64_encoded_image', 'time_of_event': 'ISO_datetime' }
Output: { 'action_taken': 'string', 'tool_invocations': [{'tool_name': 'string', 'parameters': {}}], 'explanation': 'string', 'resource_usage': {'cpu_cycles': int, 'llm_tokens': int} }
Execution harness: Docker
- Task 2
Input: { 'scenario_id': 'string', 'current_stock': {'product_a': int, 'product_b': int}, 'demand_forecast': {'product_a': int, 'product_b': int}, 'thresholds': {'reorder_point': int, 'max_stock': int} }
Output: { 'actions': [{'type': 'order_placement'|'stock_adjustment', 'product': 'string', 'quantity': int}], 'explanation': 'string', 'resource_usage': {'cpu_cycles': int, 'llm_tokens': int} }
Execution harness: Docker
Recipe freeze marker
No recipe freeze time is recorded.