Industrial safety / Canada
A complete PPE compliance platform, from labelling through to edge deployment and the web portal
We delivered the whole stack for monitoring protective equipment compliance: the labelled dataset, the trained models, the AI pipeline design, deployment onto edge hardware, and the full-stack site their safety team uses to review violations.
- Client
- Benchmark Data Solutions
- Location
- Canada
- Industry
- Industrial safety
- Engagement
- End-to-end delivery
- Period
- 2025
What we delivered
- Data labelling
- Model training
- AI pipeline design
- Edge deployment
- Full-stack portal
The problem
Compliance was being checked after the fact by people reviewing footage, which meant violations surfaced late and inconsistently.
The imagery was difficult: variable lighting across shifts, partial occlusion between workers, and protective equipment that looks similar at distance.
How we approached it
We labelled a project-specific dataset rather than adapting a generic model, working from their own site footage so the class definitions matched their compliance rules exactly.
We designed the AI pipeline end to end, then optimised the models to run on edge hardware at the site itself, which removed the dependency on a reliable uplink.
Alongside that we built the full-stack portal their safety team works in: a review queue for flagged frames, an audit trail against each one, and reporting for the compliance record.
Where it landed
Compliance review moved from ad-hoc footage checks to a consistent queue, with detection running at the site.
Because we owned every layer, there was no handover boundary between the dataset, the model, the deployment and the interface.
Highlights
- Single team across dataset, models, pipeline, edge and web
- Models running on edge hardware at the site, not in a data centre
- Web portal for reviewing and actioning flagged violations
Services involved
- Computer Vision Systems
- MLOps & Model Deployment
- Web Platforms & SaaS
- Data Services
Stack
- PyTorch
- Ultralytics
- ONNX Runtime
- FastAPI
- Next.js
- PostgreSQL
Working on something similar?
We'll tell you what we'd do differently and what we'd reuse.