In small machine shops, checking if a part can be made usually takes a lot of time. Managers pull out drawings, measure by hand, and check tool availability. This process can take 30 minutes or more per job, which adds up for busy shops handling dozens of requests weekly. Mistakes happen, leading to wasted materials, delays, and unhappy customers. To fix this, a new AI-based system was built to quickly assess part manufacturability without manual effort.

The Challenge of Traditional Feasibility Checks

Manual checks in CNC shops are slow and prone to errors. Managers must interpret CAD drawings, estimate tool paths, and verify machine capabilities. This often results in delays and sometimes accepting jobs that later turn out to be impossible to produce. When a part can’t be made as expected, it wastes time and resources, and can damage customer trust. The need for a faster, more reliable solution became clear.

The Power of AI and AMD Hardware

The new system, called MachinaCheck, uses artificial intelligence to automate these feasibility assessments. It lets users upload a CAD file, specify material and tolerance requirements, and then generates a detailed report in just 30 seconds. The core of this system is built on AMD’s MI300X hardware, which enables the AI to run entirely on-premise. This approach keeps sensitive design data secure, addressing confidentiality concerns that come with sharing proprietary geometry with third-party services.

The AMD MI300X chip has 192GB of high-bandwidth memory, allowing the AI models to operate locally without transmitting data outside the shop. This “privacy by design” ensures customer IP stays protected, making the system suitable for industries with strict confidentiality needs, like aerospace or medical device manufacturing.

The AI Workflow in MachinaCheck

MachinaCheck’s architecture is made up of five main parts, all orchestrated with a framework called LangChain and a fast API. The first part is a STEP file parser built with Python tools like cadquery. It reads CAD files directly, extracting key geometric features such as holes, surfaces, and dimensions with 100% accuracy—no guessing or image recognition involved.

Next, the system passes this data along with user inputs—material type, desired tolerances, thread specifications—to an AI model called Qwen 2.5 7B. This model, running on the AMD hardware, determines what CNC operations and tools are needed. For example, it knows that steel 304 requires carbide tools, or that a tight tolerance needs a high-precision machine. Then, a simple Python agent checks the shop’s tool inventory to see if all necessary tools are available, avoiding unnecessary AI use and reducing latency.

The third step involves another AI component that reviews the tool availability results and makes a structured decision. It considers whether the job is feasible, identifies missing tools, and suggests actions like purchasing new tools or adjusting machine settings. This decision-making process produces a clear, structured output, including confidence levels and potential risks.

Finally, a report generator compiles all the information into a professional manufacturability report. It summarizes the overall status, highlights key findings, and offers recommendations. This report helps shop managers make quick, informed decisions, saving time and reducing errors.

Running entirely on AMD MI300X hardware, this system demonstrates how AI can transform manufacturing workflows. It provides fast, private, and reliable assessments, making it easier for shops to accept jobs confidently. As a result, shops can increase throughput, improve accuracy, and maintain customer satisfaction—all while keeping sensitive data secure.

Inspired by

• https://huggingface.co/blog/lablab-ai-amd-developer-hackathon/machinacheck

Sources

• github.com