Instructor-Controlled AI in the Machine Shop: Nectir at Chabot College
Adam Hathaway grades code that can make or break things.
He's a Machine Tool Technology instructor at Chabot College in Hayward, California. His shorthand for people who've never heard the term: “We make metal smaller.”
Nearly everything in the manufactured world, from particle colliders to the embossed pattern on toilet paper, passes through some kind of machining process. Within that, he specifically teaches computer-driven manufacturing courses, i.e., Computer Numerical Control (CNC) machining, in which students learn to write the code that drives mills and lathes.
Modern machining runs on code. His students write G-code, and Adam grades it line by line. That grading is a safety check.
"A moved decimal point can crash a machine that's several thousand dollars' worth of expense. Ruin a part. Possibly cause an injury."
So when he says he needs to catch every error in a student's program, he means it literally. His grading process was slow and unforgiving, and it ate up enormous amounts of his time. That's the pressure that sent him looking at AI in the first place.
From "never touched it" to writing prompts in plain English
Nectir AI was Adam's first real experience with AI at all. He came in with no fluency and a healthy dose of “what-am-I-getting-into.”
His “aha” moment came from seeing the problems AI could solve. Nectir AI offered a first pass for both him and his students. He could get grading relief from a Nectir AI Course Assistant grounded in exactly what he looks for in a program. For students, rather than bringing the same questions to Adam, they can ask the Assistant, which will direct them on how to find the right answer rather than just handing it over.
Another surprise to Adam was how it felt to build his Nectir AI Assistant. Adam comes from a programming background, where you write very specific code to get very specific outputs, and the process is opaque.
"It was thrilling to be able to write something in just conversational English, and then to have that go through this computer process and get very specific outputs. Without having to learn code."
His fears got put to rest almost immediately. Once he saw it could work, the job became massaging the prompt until the output matched exactly what he wanted.
Building Assistants like carving stone
Adam didn't build one Nectir AI Course Assistant. He built a separate one for every assignment because each assignment has different rules, and keeping them separate kept things clear for the AI about what it should and shouldn't do.
The first setup was quick. Then came the editing, which remains a live, iterative process, though his edits are now minor.
"You start off with the big pieces. You knock off all of the major things that you know are not going to be in that sculpture, and now we're kind of down to the sanding."
The hardest part to get right was also one of the most important: keeping the Assistant from just giving students the answers. It's a tricky thing to ask of an AI. You hand it all the correct information, then tell it to help without revealing any answers.
"I had to become very surgical in how I phrased the prompt so that it wouldn't overshare information. And more importantly, couldn't be tricked by the students to give it the answer."
Chabot College teaches on two different machines that speak different dialects, so the first thing the Assistant does is figure out which program the student submitted and then check it against the correct master. When something's wrong, it flags the exact line with a note right there, so a student can go straight to the fix instead of hunting through a disembodied list of errors.
And it answers in Adam's terms. Ask it to calculate an RPM, and it returns the Chabot equation, the one he actually teaches.
"It's really important that it's my knowledge and not just search-the-internet knowledge."
That distinction is the difference between an Assistant that reinforces the course and one that contradicts it. Other AI tools would hand a student a slightly different formula and a slightly different answer, close enough to be wrong for the assignment.
The AI is designed to be flexible too. As Adam says, students don't all drill holes in the same order. One goes clockwise, another counterclockwise, and both can be correct even if neither matches the master exactly. Adam built the Assistant to follow a sensible path while still catching students who are off course. His strict prohibitions list even bans the Assistant from using phrases like "should be" or "correct value is," so it can't leak an answer through the back door.
"When it works, it just looks like it's the easiest thing in the world."
Guardrails at the center
Students first tried to jailbreak the AI Assistant. They hunted for the prompt that would make it surrender an answer, treating it like every other AI tool that could do their homework for them.
It didn’t work.
Adam could tell from his student’s submitted program that it was "a strained AI response" and told the student directly that there was no way around it but to learn the material. He calls this satisfying on two levels: he got to have a real conversation with the student about the material, and it confirmed that his prompt engineering held up under adversarial pressure.
More broadly, Adam believes the Assistant has kept students more engaged. A student working on a project at night used to hit a wall and either wait or quietly fall behind. Now there's a resource on the other side of that wall at any hour.
“Having a knowledgeable resource that they can pose questions to and get immediate answers back 24/7, I feel like that has actually kept a lot of students more engaged in the course that might have sort of fallen off.”
Adam went through this exact program himself as a student, so he knows the time cost of not having Nectir AI versus having it accessible to students now.
"One line of code can have you going through 5 or 6 different paper references for you to try to figure out what each value in that line of code should be… It speeds up student success because they don't spend time shuffling paper."
Beyond speed, there’s a sense of comfort and support in having an “expert” in the field available 24/7 to check your responses.
The Chabot case study puts a number to the scale of it: more than 5,300 lines of student code were reviewed and graded using instructor-controlled AI. Every one of those still runs on Adam's rules.
Through all of it, the reason he trusts the tool is the same reason he built it the way he did.
"I trust Nectir AI in my course because I am the one who is in control of the knowledge, the prompt, the computing, and the output. And that it continues to succeed at meeting my expectations."
He also makes a point of being polite to his Assistants, just in case they end up running the universe someday. Smart hedge. If they do, they'll remember Adam was cool.
Want to see what AI support looks like when it's built into the tools your campus already uses and controlled by your faculty? Schedule a demo, and our team will walk you through how it works at schools like yours.
Frequently asked questions about Nectir AI
What is Nectir AI? Nectir AI gives colleges, universities, and high schools the ability to deploy AI Assistants across their campus that are fully controlled by faculty and administrators, built into existing learning management systems, and compliant with FERPA and SOC 2 Type II standards. Nectir is trusted by more than 130 schools and 150,000 students and educators worldwide, including a partnership with the California Community Colleges, which serves 2.2 million students across 116 campuses.
How do students access Nectir? Nectir is built directly into the learning management system, so there's no separate login or app. Students open their course page, and the Assistant is right there, available 24/7.
How does Nectir keep AI from becoming a shortcut? Faculty control the guardrails, so an Assistant can guide students toward an answer without handing it over, and it draws only on the instructor's materials rather than the open web.