AI-Powered Hazard Detection for Heavy Industry
Industry
AI
Company size
11 - 50 Employees
About
Presien is a Sydney-based physical AI company that transforms heavy industrial machinery and worksites into intelligent assets using advanced AI vision. Spun out of the global construction firm Laing O'Rourke in 2020, its mission is to eliminate workplace hazards and improve safety without altering standard operating procedures.
Reporting Product
Figma prototypes across eight core surfaces — Summary Data, Action Centre, Blindsight Index, Review Detections Workflow, Map View, Gallery View, Charts and Data, and My Library — sufficient to move to build.
The Situation
Presien (formally PFP Robotics) builds Blindsight — an AI-powered hazard detection system for heavy industry. Cameras and sensors mounted on heavy equipment detect dangerous situations in real time: pedestrians entering exclusion zones, proximity incidents between machines and people, PPE non-compliance. In mining, construction, and logistics, these aren't edge cases. They're the daily operating environment where serious injuries and fatalities happen.
By late 2022, Blindsight was working. The AI was detecting hazards. Enterprise clients — among them Laing O'Rourke and other major contractors — had the hardware deployed on site. Data was flowing.
And nobody could make sense of it.
Blindsight was generating a significant volume of detections — camera footage, timestamped hazard events, location data, severity classifications. But the interface through which clients were meant to review, understand, and act on that data didn't exist in a form that was usable at scale. The data was there; the comprehension layer wasn't.
Kieran Mackenzie (Founder/CRO) and Daniel Moore (CTO) had been thinking about this problem through the second half of 2022. What they described to Halcrow in December 2022 was a two-part challenge: enterprises needed to be able to review the detections Blindsight was generating (triaging what was a genuine hazard versus a false positive), and they needed to be able to report on what the data was telling them across their operations.
The product concept they landed on, in Kieran's words: "think Tinder swiping for safety detections" — an efficient workflow for reviewing detections quickly — combined with safety reporting dashboards that gave site managers, safety officers, and executives a meaningful view of what was happening across their operations.
A Figma prototype had been started by Laing O'Rourke's team previously. It pointed in the right direction but wasn't quite what Presien needed — useful as a visual reference, not a finished design.
Why they called us
Kieran reached out to Halcrow through the Sydney deep tech ecosystem. The engagement grew from an initial meeting in December 2022 where the team saw Blindsight hands-on at Presien's office in Redfern — a short walk from Halcrow's own studio.
The requirement was specific: Presien didn't need engineers yet. They needed product designers who could take a well-defined problem — too much data, not enough comprehension — and translate it into interfaces that their analogue, operationally-focused users could actually use.
Kieran was direct about the user context. Their customers — site managers, safety officers on major construction and mining projects — were not digital natives. "Clean and clear — lots of whitespace and explanations — is important." The design needed to work for people who'd spent their careers on sites, not in front of screens.
Sam's response to the scope document, after reading it carefully: the requirements were mostly clear. The design system, primary user stories, and prototypes needed a Lead Product Designer's attention (Ciara Beresford, 9 days), with supporting designers on secondary user stories and usability testing (Naomi Li) and component/library work (Selena Wu). Sam would handle product management and user research strategy. Yousif Hassan, delivery management.
Halcrow pushed for one change to the scope: usability testing should include Halcrow's UX researchers in the room alongside Presien, not just designers working independently. The reasoning: "what users say they do and what they actually do are mostly not the same." Kieran agreed — with some selectivity about which customers to involve, given his view that early-stage customers often can't articulate what they need.
The MSA was signed. Work commenced.
Law 1: Start with the objective, not the solution. The design objective here wasn't "build a dashboard." It was "help an enterprise safety officer understand what their AI system is telling them, and act on it quickly." Those are different problems with different design answers.
How we worked
Embedded Design: Eight Prototypes Across Two Sprints
The engagement was structured around sprint cycles with regular check-ins and budget consumption reviews — Kieran's preference was $10k check-ins, giving both parties the ability to adjust scope as understanding deepened.
Presien's offices and Halcrow's studio were close enough that face-to-face iteration was practical. Kieran made himself available for rapid design reviews throughout January.
The design challenge — eight surfaces:
Summary Data: The landing view for an enterprise user — what happened on site today, this week, across the portfolio. The design question: what's the right level of aggregation for a safety officer who needs to understand patterns, not individual events?
Action Centre: The priority queue for items requiring human attention. Not all detections are equal — some require immediate response, some are informational. The workflow needed to surface the right things at the right time without overwhelming the user.
Blindsight Index: A proprietary Presien metric aggregating safety performance across a site or fleet. The design challenge: making a composite score legible and trustworthy to users who are sceptical of black-box AI.
Review Detections Workflow — the "Tinder" interface: The core review mechanism. Operators needed to move efficiently through detections — confirming genuine hazards, dismissing false positives, flagging for follow-up. Speed and accuracy both mattered. Swipe-style interaction patterns were explicitly referenced as the model; the design needed to make this feel fast and low-friction on a desktop interface.
Map View: Spatial representation of detections — where on site events were occurring. Exclusion zone visualisation using geospatial coordinates, overlaid on site maps. Sam noted the Laing O'Rourke prototype's exclusion zone visualisation was clever; extracting lat/long coordinates and plotting to a map service was the implementation approach.
Gallery View: Visual browsing of detection footage. Safety teams need to see what happened, not just read a classification. The gallery needed to make video and image review efficient at volume.
Charts and Data: Time-series and comparative reporting. How is this site trending? How does this fleet compare to last quarter? The design needed to serve both the site-level safety conversation and the executive-level portfolio view.
My Library: A personal saved-views and reference library for users to store the analyses and reports they return to regularly. (Notably: without a "Favourites" button — Kieran was specific on this.)
Design system and brand alignment: Presien's brand assets were provided. Their brand guide included distinctive data visualisation graphic work — circular and geometric forms used to represent site data. The design system needed to reflect this. Kieran noted their website didn't yet reflect the brand guidelines; the product design would lead the brand expression, not follow a web presence that was "frankly rubbish" and being redone.
Mobile responsiveness decision: A deliberate product decision was made — not a design limitation. The volume and complexity of data in the Blindsight reporting interface wasn't appropriate for a mobile screen. The design implemented a responsive "wall" that would detect small-screen access and redirect users to desktop or laptop. Mobile was scoped as future work.
Commercial and support data: Presien had additional data categories beyond safety — commercial data (machine hours, CO2 emissions, asset tracking) and support data (technical diagnostics for Presien's own team and distributors/OEMs). Both were explicitly out of scope for this engagement, with clear definitions of what each covered. Safety was the priority.
WHAT CHANGED
The engagement delivered the Figma prototype suite across all eight surfaces — the comprehension layer Presien needed to take Blindsight from a data-generating AI system to a platform its enterprise clients could actually use to manage safety.
By May 2023, Presien was announcing Blindsight Enterprise Reporting publicly — the sneak peek in their product newsletter showing the reporting capability that had been designed through this engagement now moving toward build and release.
For enterprise clients who had deployed Blindsight hardware on site, the reporting layer changed the commercial proposition fundamentally. Previously: AI is detecting hazards on your site. Now: here is what your AI is telling you about your safety performance, in a form you can brief your board on, your site managers can act on daily, and your safety officers can use to identify where risk is actually concentrated.
WHY THIS WORKED
AI systems that generate data without giving their users a comprehension layer are harder to justify commercially and harder to trust operationally. The investment in building something technically sophisticated doesn't translate into value unless the people responsible for acting on what the technology tells them can actually understand it.
Presien understood this clearly. The product brief wasn't "make a dashboard" — it was "solve the comprehension problem for our enterprise clients." The design work was oriented toward that objective throughout.
Law 1: The objective is what the client's customer needs to be able to do. Presien's clients are safety officers and site managers who need to run safer operations. The design worked backward from what they needed to understand and act on, not forward from what the Blindsight system was producing.
The user context also mattered. Kieran's description of his users — "pretty analogue and simple, clean and clear is important" — wasn't a limitation to design around. It was the design brief. Interfaces for operationally-focused professionals in heavy industry need to communicate clearly, surface the most important things first, and get out of the way. That's harder to design than complex dashboards. It requires real restraint.