About HardHat Pulse
Safety systems that only activate after an incident don't prevent injuries. We built the tool to change that.
Built after watching a preventable fall
Jake Morrison spent six years as a safety engineer on commercial high-rise projects in Houston before co-founding HardHat Pulse. The specific incident that pushed him to act: a roofer on a $180M medical center project fell from an unsecured scaffold in 2021. The camera footage showed the scaffold had been set up incorrectly four hours earlier. No one had walked that zone since the morning inspection.
The traditional solution — more frequent walk-throughs — doesn't scale. A 10-acre jobsite with 300 workers can't be manually surveilled around the clock. Jake and his co-founder, who had spent 5 years building computer vision systems for autonomous vehicle safety, spent 18 months building a system that could.
HardHat Pulse launched its first pilot installation on a Houston industrial build in Q3 2023. The first OSHA 300-eligible near-miss it detected was on day four of the pilot — a worker without a hard hat entering a crane swing radius. The site superintendent confirmed it would have been missed by the next scheduled inspection.
Where we've been
Incident & Idea
Preventable fall on a Houston project triggers research into persistent vision-based monitoring.
Prototype Built
First YOLOv7-based PPE detection prototype tested on archived jobsite footage. 91.2% accuracy at initial benchmarks.
First Pilot Live
First paid pilot installation on active Houston industrial site. 8 cameras, 40 wearable clips. First near-miss detected on day four.
Seed Round Closed
Closed Seed Round. 12 active pilot sites across Texas and Louisiana. Procore integration launched.
What sites report after deployment
Data from active pilot installations. Client names withheld; industry and project type included.
Industrial Facility Expansion, Texas
A $220M industrial plant expansion project in Southeast Texas recorded a 43% reduction in recordable incidents across the first 6 months of deployment. Site safety manager attributed the change primarily to the hardhat non-compliance alert cadence, which drove behavioral change in two high-risk sub-contractor crews within the first two weeks.
Commercial Tower Project, Louisiana
A 28-story commercial tower project in New Orleans detected 87 zone boundary violations in the first 30 days — 73 of which occurred between the 4:30 PM and 5:00 PM window. The GC adjusted end-of-shift protocols based on this data. No zone violations occurred in the same window during the following 45 days.
Highway Bridge Rehabilitation, Texas
On a TxDOT bridge rehabilitation project, a fall-detection alert from a worker's wearable sensor reached the site nurse 11 minutes before the worker was discovered by a co-worker on the far side of the deck. The worker had a minor ankle fracture. The faster response was cited in the incident report.
How we work
Field-first design
Every feature is validated on active construction sites before it ships. If a safety supervisor can't use it in a hard hat and work gloves on a loud jobsite, it doesn't meet the bar. We do not build for conference demos.
Privacy by design
Construction video doesn't leave the site unless an alert clips it for incident review. We don't use client footage to train models without a signed consent agreement. Worker tracking data is pseudonymized at the edge — we see tag IDs, not names.
Honest about accuracy
We publish F1 scores and false positive rates from production deployments, not lab benchmarks. If the model performs below threshold on a site, we say so and fix it before billing starts. The 30-day pilot exists specifically to prove accuracy in your specific conditions.
Meet the team behind the platform
Three people who've worked construction sites, built computer vision systems, and decided to fix the thing that bothered them most.
Meet the Team