Using AI to Enhance Your Machining Experience

If you're like us, we use AI tools like ChatGPT or Claude daily to help streamline our tasks around the shop and find it genuinely useful in most cases. But sometimes, we step back and wonder "where the hell did that come from... that's just wrong!" It turns out that AI hallucinations are a real thing, and when you're working with expensive tooling on six-figure machines, you don't want bullshit answers.

One of the best tools we've found to save us time around the shop is Google's NotebookLM. NotebookLM is Google's AI research assistant designed to work exclusively with documents you upload (PDFs, Google Docs, Word files, ePubs, web links, and more), making it especially useful for digging into PDFs of Manuals, Data Sheets, Tooling Guides, etc. You upload your "sources" into a notebook, and the AI essentially becomes an instant well-read 'expert' on that specific collection of material rather than drawing from the open internet.

The way it reduces hallucinations comes down to two main techniques:

Source grounding via RAG. NotebookLM pairs Google's Gemini model with a retrieval-augmented generation pipeline or 'RAG'. RAG segments documents, creates semantic embeddings, and retrieves relevant passages before generating an answer. If the information isn't in your uploaded files, the model is supposed to tell you it doesn't know rather than fabricate something. We've found this to be true the vast majority of the time.

Inline citations on every answer. Responses include citation chips that link directly to the specific passage in your document, so you can verify every claim against the original PDF in one click. This is arguably the most valuable feature as it cuts down the time it takes to get to the necessary information you need to do the task at hand.

Below is an example of how Versicor Precision uses GoogleLM with our Mazak Integrex i250HS:

We upload all of the reference manuals that we received with the machine, errata or updates from Mazak, Service Bulletins, and any other official documentation we have into NotebookLM. We also maintain a document of 'Lessons Learned' that contains any tips and tricks we've learned over time to ensure that we don't make the same mistake twice.

In the example above, we were programming a part that needed the W sub spindle to traverse over with a Royal Products Ultimate Multi-Tasking Live Center to support the stock while doing some 5-Axis work, and wanted to make sure that we could shut the machine down overnight if a part was in process. Normally, Mazak requires all axes to be homed at power-up, so we wanted to disable that requirement.

We asked the LM "What parameter disables the W axis homing check at powerup" and it immediately instructed us to check RB21-Bit 5 and provided a link to the manual and section needed to verify, along with other related parameters to check.

It took longer to open the browser than to locate the precise settings we needed to change! Previously, we would spend 5-10minutes trying to remember which of the 37 PDF documents we needed to look in to find parameter settings, and another 5-10 interpreting the manuals.

We still "Trust But Verify" any interactions with any AI tools, NotebookLM included, but we find that the time saved is a measurable benefit to our bottom line and is well worth the effort.

We hope this glimpse into one of our processes is helpful and helps you in your CNC journey!