Skip to main content
Get started
Our Principles

Ethics & Principles

Building AI for safety-critical industries means thinking honestly about what can go wrong, not just what can go right.

I'm a mechanic and supervisor with more than 23 years' experience on heavy equipment across Australia and Canada. I built TroubleshootHD because I've lived the problem — watching good techs waste time hunting through manuals, watching juniors stand around waiting for a supervisor callback, watching institutional knowledge walk out the gate every time someone quits.

But building an AI tool for this industry means thinking honestly about the trade-offs. This document lays out how I think about the ethical dimensions of this platform, what I'm doing about them, and where I'm still figuring it out.

These are the principles shaping how I build the platform and where I draw the line..

Core Principles

Six Rules We Build By

01

Make Technicians More Capable, Not More Dependent

TroubleshootHD should make a mechanic better at their job — not unable to function without it. There's a critical difference between a tool that compresses the time between "I don't know" and "now I have the information to figure it out," versus a tool that does the thinking for them.

What this means in practice
●Answers drawn from your manuals cite the manual and page, and direct techs to the relevant section rather than just giving a standalone answer
●The platform surfaces information in context — the tech still has to interpret it, apply judgment, and do the work
●We will never position TroubleshootHD as a replacement for trade training, apprenticeship programs, or hands-on learning
●Learn Mode exists to build understanding of systems, not just deliver lookup results
02

The Wrench Still Has to Turn

AI cannot verify what's physically in front of the technician. It doesn't know if a machine has been field-modified, if a component has been substituted, or if conditions on the ground differ from what the manual assumes. The human in the shop is the final authority, always.

What this means in practice
●Wolf Vision and Learn remind techs to check the manual before acting on an answer.
●The platform supports the technician's decision-making, not replaces it
●We actively refuse to provide instructions that bypass safety devices
●We encourage techs to validate, question, and override AI-generated content based on what they see on the machine
03

Supplement Mentorship, Don't Replace It

The best mechanics didn't just know where to look things up — they had internalized mental models of how systems behave. That deep knowledge came from the friction of learning: getting it wrong, asking a mentor, building the map in your head. If a tool removes too much of that friction too early, it could produce techs who are efficient at retrieving answers but less practiced at reasoning toward them.

What this means in practice
●The platform is positioned as a complement to supervision and mentorship, not a replacement
●Marketing and sales will never suggest that shops can reduce experienced headcount by adopting the tool
●Feature design favours presenting relevant information for interpretation over delivering direct instructions
●We pay attention to whether features are helping techs learn or just shortcutting thinking.
04

Systems Should Support Knowledge, Not Replace It

Repair logs, shift handovers, and shared knowledge bases are valuable because they capture what people know. But if the platform becomes the only place that knowledge lives — and no individual person carries it anymore — the organization becomes fragile.

What this means in practice
●We encourage shops to maintain mentorship and knowledge-sharing culture alongside platform use
●Your admin can export your organization's records — no vendor lock-in on your own knowledge
●We're transparent that the tool is a layer on top of your team's expertise, not a container for it
05

Safety Is Non-Negotiable

Heavy equipment can kill people. That fact shapes every decision about what the AI will and won't do. Old Mate and Wolf Vision refuse to give instructions for bypassing safety systems.

What this means in practice
●"I can't provide instructions to bypass safety devices — that's a line I won't cross."
●Operational/diagnostic overrides (DPF regen, idle timers, diagnostic mode) are allowed — they're routine troubleshooting
06

Data Ownership Is Clear

Organizations own their organizational data. Individuals own their personal data. There's no ambiguity. Multi-tenant architecture enforces strict data isolation between organizations via Row Level Security.

What this means in practice
●Organizational data (repair logs, equipment records, manuals, handovers) belongs to the organization
●Personal data (chat history, bookmarks, preferences) belongs to the individual user
●If a tech leaves a company, their personal data goes with them; the org's data stays
●We don't sell user data, we don't train external models on customer manuals, and we don't share data between organizations
Honest Trade-offs

Tensions We Think About

These aren't reasons not to build the platform. They're tensions we hold in mind and design around.

The Knowledge Formation Question

When you give a junior tech instant access to answers, you risk producing a generation that's great at retrieval but weaker at reasoning. The best mechanics have internalized mental models built through years of friction.

We design the tool to present information in context rather than as standalone answers. Cite sources so the tech reads the manual section, not just a summary. Build Learn Mode features that develop system-level understanding. Stay close to beta users and watch for signs that the tool is replacing thinking rather than supporting it.

The Labour Market Shift

If TroubleshootHD is widely adopted, one experienced supervisor can effectively oversee more junior techs. The economic incentive for shops is to hire more juniors and fewer seniors.

The platform is positioned as a tool that raises the floor — not one that lowers the ceiling. The techs who thrive in a tool-augmented world are the ones who combine information access with genuine mechanical intuition. We make the case for investing in good technicians, not for replacing them.

The False Confidence Risk

Better information access means fewer mistakes. But there's a risk of overconfidence: a tech who trusts an AI-cited answer without validating it against the physical reality in front of them.

Safety disclaimers aren't just legal cover — they're a design choice. Wolf Vision and Learn remind techs to check the manual before acting on an answer. Cited sources let them check the original material. The culture we build around the tool matters as much as the tool itself.

The Institutional Knowledge Paradox

If the platform becomes the single repository of institutional knowledge and people stop carrying it in their heads, the organization becomes dependent on the platform.

We're explicit that TroubleshootHD is a layer, not a replacement. We encourage knowledge-sharing culture alongside platform use. Data export ensures organizations aren't locked in. But we acknowledge this is a tension inherent in any knowledge management system.

"Mechanics and supervisors in mining and construction have some of the best bullshit detectors in any profession. They've sat through enough vendor pitches to know when someone's overselling."

TroubleshootHD does some things really well and it also has limits. I'd rather be upfront about both than pretend otherwise. If you've read this and have thoughts — as a user, a prospect, or just someone in the industry who cares about where this is heading — I want to hear from you.

This is a living document and it gets better with input from the people who do this work every day.

— Jarrad, Founder. Mechanic and supervisor with more than 23 years' experience. Still learning.