Meet Cason Shepard! He’s an AI engineer at Axle who became a VMRS specialist because he wanted to understand the nitty gritty work behind the data.
That curiosity extends well beyond VMRS, and influences how we build products at Axle every day. Cason is also an excellent baker (vegan chocolate chip cookies = better than Tate's), very casually ran the SF Marathon over a random weekend, is so much fun to be around, and continues to make a heroic - if futile - effort to keep our office fiddle leaf fig alive (RIP Mr. Fig) We know he’s nearby when we hear a signature “What the heck!” We're v lucky we get to work with him!!!
Meet Cason
AI engineer + VMRS specialist
Cason builds AI for repair and maintenance. He’s also the person who’ll happily spend an afternoon digging into VMRS codes, repair workflows, and out-of-service issues. For him, knowing those details is how you build tools fleet and service teams can actually rely on.
Most AI engineers never go anywhere near VMRS. What made you want to learn it?
VMRS can encode data at every level of specificity: what system of the vehicle was worked on, what repair actions were performed, all the way down to a single component. As someone who has always enjoyed intuitive systems, graph theory, and mnemonic devices (PEMDAS!!), I find it the perfect intellectual playground.
It’s also a perfect fit for AI. It’s a well-defined taxonomy with clear rules, it generates lots and lots of data, and fully using it takes constant upkeep that no human can realistically keep up with.
What does AI get wrong about maintenance data when the person building it doesn’t understand the specifics?
Fleets and shops run their organizations in specific, specialized ways, like unique internal ID systems or separate internal and external work order systems. There’s no one-size-fits-all AI for that.
You can’t just use AI to “read an invoice” without first understanding why you’re reading it. Are we catching overbilling? Logging service dates? How do certain providers bill differently? These goals shape how AI systems should be built, not just how they’re interacted with.
On predictive maintenance
Predicting when a vehicle will break down is genuinely valuable, but it’s often treated as the fix when the root issue is simpler: preventive maintenance schedules that aren’t being tracked well.
Fleet managers know this. When a truck misses its scheduled maintenance, it’s much more likely to break down. That’s why I enjoy building AI-native workflows that work alongside a fleet manager’s extensive working knowledge rather than trying to replace it.
How do you think about the technician, fleet manager, or service writer on the other end?
In any job there are a ton of small things we all do to keep the lights on, like filling out a form or sending a quick update.
Fleet managers and service writers do plenty of this, while also overseeing work that requires real care: their fleet’s health and a vehicle’s safety on the road.
Our job is to build systems that harness not only AI but the expertise our users already have. Only once we understand how they work day to day can we find the right place for AI in their workflows.
View the original slides
