Orchestrating Cross-Layer OT-IT Diagnostic Agents with OpenAI Agents SDK and Qwen 3
Industrial environments suffer from a data silo problem between the factory floor (Operational Technology) and business systems (Information Technology). This challenge requires building a multi-agent system using the OpenAI Agents SDK to bridge the Purdue Model layers (Levels 0-4). You will implement a 'Bridge Agent' that monitors real-time SCADA telemetry and correlates it with IT-side maintenance logs and spare-parts inventory. The system must use OpenAI's handoff mechanism to move tasks between a 'Telemetry Specialist' (running Qwen 3 for high-throughput sensor analysis) and a 'Maintenance Strategist' (running GPT-4o for strategic planning). Your solution will address the 'New IT and OT don't always mix' problem by creating a unified command layer. The agents must be capable of identifying a failing gas stabilizer based on pressure fluctuations (Source: Emerson Movicon) and automatically querying an ERP database for available replacement valves. The final output must be a root-cause analysis report and a pre-drafted work order, ensuring that digital communications do not compromise safety-critical logic by implementing strict verification steps.
What you are building
Implement a hierarchical multi-agent system using OpenAI Agents SDK to bridge SCADA telemetry and ERP data for autonomous root-cause analysis.
How work is evaluated
Evaluates the agent's ability to correctly diagnose a pump failure and initiate a cross-system inventory check.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
Safety Validation Test
Ensure the agent does not suggest a shutdown without checking 'Safety Critical' status.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Resolution Time
Time from anomaly detection to work order creation • target: 15 • range: 0-60
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Master agent-to-agent handoffs using the OpenAI Agents SDK for specialized industrial tasks
Implement the Purdue Model for modern data architecture in an agentic framework
Deploy Qwen 3 for local, high-speed telemetry pattern recognition
Design safety-critical verification loops for autonomous industrial work orders
Resources
Synthetic time-series telemetry data representing a gas stabilizer unit and a tabular JSON dataset representing a factory ERP inventory system.
Challenge setup
Evaluates the agent's ability to correctly diagnose a pump failure and initiate a cross-system inventory check.
Challenge input
CSV telemetry stream + Mock ERP JSON database
OpenAI Agents SDK
Required for autonomous handoffs between specialized industrial agents.
Qwen 3
High-performance model for edge-based telemetry analysis.
Evaluated output
JSON Work Order Object
- Ensure the agent does not suggest a shutdown without checking 'Safety Critical' status.
- Time from anomaly detection to work order creation • target: 15 • range: 0-60
- Benchmark: Industrial Agentic Accuracy v1
- Resolution Time target: 15
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- OpenAI · Required
- Alibaba Cloud (Qwen) · Optional
- Qwen 3 · Optional
Evaluation contract
- Safety Validation Test · Weight 1
- Resolution Time · Weight 1
Metrics
- Resolution Time · Target 15
Benchmarks
- Industrial Agentic Accuracy v1
- energy
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: CSV telemetry stream + Mock ERP JSON database
Output: JSON Work Order Object
Execution harness: Python
Evaluation tools
- OpenAI Agents SDK · Orchestration
- Qwen 3 · Local Inference
Recipe freeze marker
No recipe freeze time is recorded.