AI Engineer

Work Location – Hybrid role in Sandy, UT Position Overview

We’re looking for an AI Engineer who gets genuinely excited about solving real problems with AI. Not someone who just wants to chase the latest models because they’re new and shiny—we want someone who asks “what will actually make workers safer?” and “what will help safety managers prevent incidents before they happen?”

This isn’t demo work. You’ll be shipping features that construction superintendents use every morning to figure out where they need to be. Features the frontline needs in emergencies. Features that Spanish‑speaking crews use to report what they’re seeing in their own words. Your work directly impacts whether someone goes home safe at the end of their day.

Leading our workflow architecture

• We need to migrate our AI workflows to N8N and build something that’s actually robust and scalable, not the patchwork of tools we’ve been using.
• You’ll design the workflow orchestration that connects our various AI models, data sources, and business logic in a way that makes sense.
• Build workflows that are observable and self‑healing, because they need to scale with us as we grow quickly.

Making our NLP better

• Our conversational AI lets workers create safety reports just by talking—in English or Spanish. You’ll optimize and fine‑tune that experience.
• We have this feature called “Ask Mojo” that turns complex safety manuals into natural conversations. It needs to get better at making sure workers get the right answer from the right document, every time.
• Build context‑aware understanding that knows the difference between “fall protection” on a 30‑story building versus a 6‑foot ladder—because that context really matters.

Pushing our OCR and computer vision forward

• Our Flex PTP technology extracts structured data from any safety form, whether it’s a pristine PDF or a mud‑stained piece of paper photographed at sunset on a jobsite. We need to make this even better.
• Improve accuracy on handwriting recognition, checkbox detection, form field extraction—all across wildly inconsistent formats.
• Build intelligence that doesn’t just extract text but actually understands what it means in a safety context.

Collaborating across the team

• Work with product managers to turn user pain points into technical solutions that actually solve problems.
• Partner with other engineers to make sure your AI models integrate smoothly into the broader platform.
• Talk directly to customers sometimes to understand their challenges and make sure what you’re building actually works for them.

You’ve actually shipped AI to production

• You know the difference between a model that’s 95% accurate in testing and an 85% accurate model that users actually trust in the real world.
• You’re comfortable across the full stack—from training models to designing APIs to orchestrating workflows.
• You write code that other engineers want to maintain (or at least don’t that’maintaining).

You care more about impact than the technology itself

• You get excited about solving a real problem imperfectly rather than building a perfect solution to the wrong problem.
• You measure success by what users can do, not just by model metrics.
• You’re totally fine using “boring” technology if it’s the right tool for the job.

You’re a self‑starter who can handle ambiguity

• You won’t wait around for perfectly specified requirements. You talk to users, figure out the core problem, propose solutions, and execute.
• You can make technical decisions with incomplete information and then adjust as you learn more.
• Someone can give you a vague goal like “make our OCR work better on handwritten forms” and you’ll turn that into a concrete plan with actual milestones.

You obsess over the right details

• You care about latency because a 2‑second response time is the difference between a worker using your feature or avoiding it completely.
• You think deeply about error handling because a confusing error message on a construction site isn’t just annoying—it creates real safety risk.
• You design for actual conditions: poor lighting, people wearing gloves, limited connectivity, multilingual users.
• You can explain…