XR Simulation is a mixed-reality safety-training prototype for new industrial employees, built by Hitachi Research & Development for manufacturing stakeholders. Trainees practice hazard recognition, navigate facility layouts, and rehearse emergency responses before entering a real site.

Designed and built it in Unity for Meta Quest. One of six immersive training prototypes owned from concept through stakeholder pilot. I led instructional design and Unity development.

New employees and trainees in industrial settings: oil and gas, manufacturing, warehousing, and other large-scale facilities. Most are new to the specific site. Many are new to the industry.

Give workers a safe place to practice safety procedures, learn complex environments, and respond to risk, before any of it carries consequences.

• Mixed-reality training simulation built in Unity with C#

• Developed for Meta Quest

• Modular 3D training environments based on real facility layouts

• Gamified hazard identification and safety drills

• Real-time AI dialogue avatars with spatial audio

• Portable prototype for off-site delivery

• Built with subject-matter expert input from engineering teams

Industrial safety training is expensive and hard to scale. Programs depend on physical mockups, on-site supervision, and limited access to real spaces. All of which cost money and don't scale.

The result is a gap. New workers receive instruction, pass the assessment, and still arrive at a complex facility unprepared. Knowing a procedure is not the same as having done it somewhere.

The challenge was to build familiarity and confidence before the first shift, without the cost of a physical training facility. 

I treated this as a learning design problem, not a technology demonstration. Lectures and manuals transfer information. What Marc needed was a place to practice.

Physical mockups limit how often anyone can rehearse. A portable simulation removes that ceiling. Trainees can repeat a procedure until it's automatic, at no marginal cost per repetition, without occupying a working facility.

The simulation used experiential learning. Trainees moved through the space. They identified hazards. They interacted with equipment. They practiced emergency response in context, in the order it would actually happen. 

The scenarios focused on:

• exploring facility layouts

• identifying environmental hazards

• interacting with equipment

• rehearsing evacuation and safety procedures

• reinforcing correct actions through feedback and scoring

Spatial familiarity and situational practice are both strong predictors of safety performance. Neither transfers from a slide deck.

Because the subject was safety, realism was a requirement, not a finish. We based facility layouts on real environments where possible. Engineering subject-matter experts shaped the scenarios and validated the hazards. Equipment interactions reflected real workflows, including the awkward parts.

• A portable mixed-reality training prototype

• Simulated industrial environments modelled on real facilities

• Hazard identification exercises with feedback and scoring

• Emergency response and evacuation drills

• Interactive equipment-based training moments

• AI-supported dialogue with spatial audio

On the conversational avatar. I built a real-time AI dialogue avatar so trainees could ask questions in the environment rather than breaking out of it to find an answer. The first version had the problem every early LLM integration had: it answered confidently and sometimes wrongly, which in a safety context is worse than not answering at all. I restructured the prompts and grounded the model in domain-specific manufacturing data, which substantially reduced hallucinated and off-topic responses.

I built a real-time AI dialogue avatar so trainees could ask questions in the environment rather than breaking out of it to find an answer. The first version had the problem every early LLM integration had: it answered confidently and sometimes wrongly, which in a safety context is worse than not answering at all. I restructured the prompts and grounded the model in domain-specific manufacturing data, which substantially reduced hallucinated and off-topic responses.

This project was designed to make safety training more practical and repeatable. It gave learners a way to become familiar with complex spaces and procedures before facing real-world risks.

The prototype also explored how mixed reality could reduce dependence on physical mockups. It supported more flexible off-site preparation. By combining simulation, interaction, and spatial learning, the project showed the value of immersive training in high-risk environments.

Knowing a procedure isn't the same as practicing it. In industrial settings, that gap is where incidents happen, and reading cannot close it.

 Mixed reality connects instruction to action, space, and decision-making, which is the whole point. Not that it's immersive, but that it lets someone be wrong somewhere with survivable consequences.

 The most useful thing I took from the build was how much of the design work was about restraint. Every interactive element competes for a trainee's attention, and attention spent on novelty is attention not spent on the hazard. The best version of this training is the one where the technology stops being interesting.