Explore MCP
Try out MCP Connectors in Claude.ai
Presien reduces critical safety events on construction sites by 70%+ with Claude
.svg)
Industry:
Software
Company size:
Startup
Product:
Claude Platform
Location:
Asia Pacific
Over 70% reduction in critical safety events
within the first three months of deployment
From hours of manual review to proactive site briefings
Safety managers receive actionable AI-generated risk and site analysis
Presien is a physical AI company that runs its own computer vision models directly on heavy machinery in construction and mining, detecting safety hazards in real time. After six years building its proprietary data infrastructure, the company connected it to Claude through MCP to extend /loop, an agentic intelligence platform that reasons over live worksite data around the clock and surfaces risks before they become incidents.
With Claude, Presien achieved:
- Over 70% reduction in critical safety events within the first three months of deployment
- Safety managers shifted from hours of daily video review to starting each day with an AI-generated risk briefing, enabling them to act immediately on issues
- 24/7 autonomous monitoring across thousands of machines and millions of data points, with no manual review required
- Compliance monitoring moved from periodic manual checks to continuous, policy-aware analysis
- Supervisor briefings generated from real incidents observed hours earlier, replacing generic templates
- MCP server integration built in weeks on top of Presien's existing data infrastructure
The challenge
Explore MCP
Try out MCP Connectors in Claude.ai
Explore MCP
Try out MCP Connectors in Claude.ai
A flood of data with no way to reason over it
Construction and mining sites generate massive amounts of data, but less than 5% of it is ever acted on. The U.S. Bureau of Labor Statistics estimates that every 104 minutes, a worker dies from a preventable injury in the U.S. Operators of these sites generate enormous volumes of video and sensor data for the purposes of incident prevention and site management.
Presien spent six years building computer vision models that run directly on heavy machinery, detecting hazards and blind spots in real time from dusk to dawn. That foundation gave them something rare: a continuous stream of real-world physical data from active sites worldwide.
Before Presien created /loop with Claude, safety workflows were manual and reactive. Managers sifted through dashboards and video clips to assess risk, a process that didn't scale and left machines collecting thousands of data points when teams were already buried in review. There was no way to gain insights across a project, cross-reference alerts against regulations, or catch compliance issues before they became incidents. Reports were compiled by hand and safety briefings were often generic.
"That's what led us to Claude," said Mark Richards, CEO of Presien. "Not as a chatbot layer sitting on top of our data, but as the reasoning engine at the heart of our intelligence platform."
The solution
Choosing the right Claude model
Learn when to use Haiku, Sonnet, or Opus to get better results and stay inside your rate limit. A practical guide to picking the right Claude model.
Choosing the right Claude model
Learn when to use Haiku, Sonnet, or Opus to get better results and stay inside your rate limit. A practical guide to picking the right Claude model.
Choosing the right Claude model
Learn when to use Haiku, Sonnet, or Opus to get better results and stay inside your rate limit. A practical guide to picking the right Claude model.
Selecting Claude for orchestration across complexity
In evaluating models, what mattered most to Presien was reasoning quality across complex, multi-source data, and reliability in multi-step agentic workflows the team could trust and explain to enterprise customers.
During testing against real worksite data, Presien gave Claude a prompt: Identify recent high-risk trenching detection events, then create a toolbox talk for site crews, save a draft email to site leads, and block tomorrow morning in my calendar to review before sending. Claude executed the full sequence, reasoning across video, events, telemetry, and policy documents.
“Other models struggled orchestrating complicated tool calls, or produced output that required significant human rework,” said Richards. “With Claude, the adherence to instruction was precise, the generated content was genuinely usable, and it did all of this through our MCP server architecture, knowing which of Presien's tools to call, without explicit instruction.”
Presien provides end users with starter instructions and pre-built Skills that they add to their own Claude Project connected to Presien’s MCP server. For those using /loop, pre-built agents accelerate the journey further. Interactions are contextualized to their business: their fleet, sites, and risk thresholds. “Skills let us build repeatable workflows, like generating a branded safety report or drafting a site escalation, that anyone on the team can trigger in seconds,” Richards said. “Together, it's less like using an AI tool and more like having a team member who already knows the business.”
How /loop uses Claude and MCP to monitor sites and detect issues
Presien's computer vision models run on the machines themselves, generating a continuous stream of detection events: a worker in a blind spot, an excavator operating too close to a boundary, a vehicle moving without a spotter. Those events flow to the cloud, where Presien's vision-language model processes them and stores them alongside machine telemetry, company safety policies, and government regulations.
Claude connects to this data through MCP servers. When /loop identifies a cluster of high-risk detections on a site, Claude pulls the relevant events, cross-references them against the company's documentation, and determines what action is needed. Or a safety manager can query /loop directly: "Where are my highest risk activities across all sites this week and what should I do about it?"
/loop runs continuously, 24 hours a day, reviewing events as they happen. Critical detections trigger immediate attention; others are queued for morning review. By the time the safety manager starts their day, the analysis is ready.
"Not as a chatbot layer sitting on top of our data, but as the reasoning engine at the heart of our intelligence platform."
CEO, Presien
The outcome
70% fewer critical safety events in three months
Across early deployments, critical safety events have dropped by over 70% in the first three months. Without hours of daily video review, safety managers and site supervisors spend their time onsite with their teams, having the conversations that actually change behavior, rather than sitting at a desk watching clips.
The capability that represents the biggest leap is the end-to-end agentic workflow. A safety manager can issue a single instruction for a complex, multi-step task, and Claude executes the full sequence. These can also be saved as a repeatable agent that runs continuously, able to recognize a repeat scenario and deploy a workflow.
"I used to spend the first two hours of every day watching videos and filling in reports,” one site safety manager said. “Now I walk the site instead. The AI tells me where the problems are, and I focus on fixing them."
MCP gave the team the architecture to enforce that. "In a safety-critical industry, a hallucinated insight is potentially dangerous," Richards said.
What’s next
Presien is expanding /loop toward fully proactive operations, where safety and operations managers begin each day not by logging in, but by receiving a briefing built for that role. For safety managers, that means identified risks, drafted escalations, and the compliance trends that need attention. For operations, it means machine utilization, earthworks progress, productivity trends, and idle time that costs money.
Through MCP, Presien is also offering OEM partners—the manufacturers of excavators, mining trucks, bulldozers and other heavy machinery—the ability to embed its intelligence directly inside their own platforms, so they get a native experience without switching tools.
"The shift is from a tool you query to an intelligence that monitors, interprets, and surfaces what matters without being asked," Richards said.
"With Claude, the adherence to instruction was precise, and it did all of this through our MCP server architecture, knowing which of Presien's tools to call, without explicit instruction.”
CEO, Presien