MachinaCheck: Building a Multi-Agent CNC Manufacturability System on AMD MI300X

Manual quoting processes create production bottlenecks and increase the risk of accepting jobs that exceed shop capability, leading to wasted machine time and scrapped components. Relying on commercial cloud APIs for confidential CAD files exposes proprietary R&D data to third-party vulnerabilities, making on-premise AI processing a prerequisite for enterprise manufacturing. Implementing this architecture allows shops to scale quote volume while maintaining strict IP compliance. A primary limitation remains the reliance on high-end hardware like the MI300X, which may present a significant initial capital expenditure for smaller machine shops.

MachinaCheck shifts from manual human verification to a five-component pipeline that integrates deterministic STEP file parsing with LLM-based decision logic. By running Qwen 2.5 7B on AMD MI300X hardware, the system achieves sub-minute analysis without transmitting proprietary geometry over the internet. The following table contrasts traditional manual assessment with this new automated workflow:

One current limitation is that the model's manufacturing domain knowledge is constrained by the initial training data and requires specific shop floor inventory integration.

This week

• Audit current manual quoting workflows to identify the average time spent on feasibility analysis per job.
• Evaluate on-premise hardware requirements for hosting local LLMs to ensure data residency compliance.

This quarter

• Pilot the MachinaCheck pipeline on a subset of incoming STEP files to measure accuracy against human estimates.
• Integrate existing tool inventory databases with the Python-based matching agent to automate capacity checks.

This year

• Scale the automated manufacturability reporting across all incoming RFQs to increase shop throughput.
• Refine the decision-making agent to include real-time machine scheduling and cost-estimation capabilities.