amanvision
Back to work Case Study · 2024
04 — AmanVision

amanvision.

An AI-powered HSE platform that turns live camera feeds into real-time safety intelligence — built for the scale, languages, and regulations of the GCC.

Year
2024
Role
Product Designer
Duration
6 months
Team
Creative Director, CTO, Co-founder, HSE Manager, Engineering
Tools
Figma · Illustrator · Claude
AmanVision camera-details dashboard on a laptop — a live industrial camera feed with a heat-map risk overlay and a side panel of detected events

The rules arrived faster than the tools to meet them.

Across the GCC, workplace safety stopped being optional. Stringent, OSH-aligned HSE regulation — tied to Vision 2030 — now reaches from mandatory PPE on every active site to incident reporting, escalation, and audit-ready documentation on demand. The obligation was clear. The software to actually meet it, for this region, was not.

What changed

New mandates landed on the same manual workflows.

  • Stringent, OSH-aligned HSE regulation tied to Vision 2030.
  • Mandatory PPE compliance on every active site.
  • Rigorous incident reporting and escalation built into the day.
  • Audit-ready documentation expected whenever an inspector asks.

Four gaps between the rule and the floor.

We started by mapping how safety actually moved through a site — not the policy, the practice. Four gaps showed up on every job.

Built for somewhere else

Safety tools weren’t designed for GCC realities — the languages, the regulations, or the site conditions.

Fragmented and manual

Compliance lived across spreadsheets, chat threads, and paper. Nothing was connected, nothing was searchable.

Risk seen too late

Violations surfaced only after they’d already caused an incident — reporting, never prevention.

No regional memory

No system captured the behaviour patterns specific to a region or a site, so the same risks kept repeating.

How might we

surface a risk the moment it happens — and turn it into a record every role can act on?

Strong tools, built for another market.

Protex AI, Buddywise, and Intenseye had polished enterprise UIs and mature detection — but heavy onboarding, Western-centric workflows, and thin multilingual support. The opening was a GCC-native platform: localised, simpler to adopt, and audit-ready out of the box.

Benchmarked against Protex AI Buddywise Intenseye
StrengthsInternal · Helpful
  • Polished enterprise UI
  • Strong analytics dashboards
  • Mature rule engines
  • High detection accuracy
  • Clean information hierarchy
WeaknessesInternal · Harmful
  • Heavy cognitive load
  • Complex onboarding flows
  • Western-centric workflows
  • Poor multilingual support
  • Limited workflow customisation
OpportunitiesExternal · Helpful
  • GCC-native compliance patterns
  • Cultural & language localisation
  • Simplified compliance journeys
  • Unified end-to-end IA
  • Role-specific workspaces
ThreatsExternal · Harmful
  • Rapid CV evolution
  • Bigger players expanding
  • Fast, copyable UX patterns
  • Rising expectations for automation
  • Vendor lock-in barriers
Competitive scan The leaders proved the model worked. None of them were built for a bilingual, GCC-native compliance reality — the left column is exactly where AmanVision could win.

One incident. Five people who have to act.

A single violation touches five very different users, each entering the same event at a different point and needing a different thing from it. Role-based access wasn’t a setting — it was the architecture.

Admin

Sets up facilities, cameras, detection rules, users, and permissions.

Safety Officer

Monitors real-time events, verifies incidents, and flags false positives.

Facility Manager

Reviews alerts, adds evidence, and assigns corrective actions.

HSE Manager

Validates events, ensures compliance, and updates rules and SOPs.

Executive / Leadership

Watches dashboards, trends, and overall site safety performance.

Role-based access control

Same event, five different systems.

Every capability mapped to every role, down to create / view / update / delete. A distilled view of the full RBAC matrix — the spine the whole product was built on.

Capability Admin HSE Mgr Facility Mgr Safety Officer Exec
Facilities & areas CRUD CRUD CRUD CRUD CRUD
Cameras & zones CRUD CRUD CRUD CRUD CRUD
Alerts CRUD CRUD CRUD CRUD CRUD
Observations CRUD CRUD CRUD CRUD CRUD
Schedules & detection rules CRUD CRUD CRUD CRUD CRUD
Analytics & dashboards CRUD CRUD CRUD CRUD CRUD
User accounts CRUD CRUD CRUD CRUD CRUD
C Create R Read U Update D Delete Faded = not granted

Real-time safety intelligence for high-risk environments.

The platform splits cleanly in two: what the cameras see, and what each role does about it. The detection layer never touches a person until it has something worth acting on.

Detection

What the system sees

  • Real-time computer-vision safety monitoring
  • PPE and access-violation detection
  • Detections converted into structured events
  • Automatic severity scoring and metadata
  • Configurable zones and rules per site
Workflow

What each role does

  • Role-based compliance workflows
  • A centralised incident dashboard
  • Faster validation and escalation
  • Compliance trends and regional insight
  • Decision-ready safety reporting

From login to audit, in five layers.

The whole platform runs as one pipeline. Access resolves a role, setup defines the site, the AI layer turns footage into events, each role works those events, and everything rolls up into reporting and audit — the master flow that held the build together.

Level 1 · Entry & Access
Authentication & Role Resolution
Login with company credentialsPlatform loads the correct workspace based on role
AdminHSE ManagerFacility ManagerSafety OfficerExecutive/Leadership
Level 2 · Organization Setup
Facility & Camera Setup
Create facilitiesCreate areasCreate sub-areaAssociate cameras
Add camerasConfigure detection zones
Assign which roles see which sites
AdminHSE ManagerFacility Manager · view
Compliance Configuration
Select required PPE types
Configure unsafe-action detections
Set incident categories & severity
Define workflows for escalations
AdminHSE Manager
Level 3 · AI Monitoring Layer (Autonomous System Layer)
Live AI Detection Layer
The AI detects:
Missing PPEUnsafe actionsRestricted-zone entryFire / smoke cuesCrowd anomalies
Facility ManagerSafety Officer
Event Generation
Every detection triggers an event object with:
TimestampCamera + locationPreliminary labelSeverity scoreSnapshots
Level 4 · Event Workflow (Core Operational Engine)
Facility Manager Workflow
Verifies events with contextual footage, assigns corrective actions, and manages day-to-day safety operations.
Event queueReviewAdd evidenceAssign resolutionClose / open
HSE ManagerFacility ManagerSafety Officer
HSE Manager Workflow (Quality & Compliance Oversight)
Reviews and annotates events for audits, escalates severe cases, and refines detection rules or SOPs when recurring patterns appear.
ReviewValidate resolutionsAnnotate eventsApprove / rejectUpdate rules
HSE ManagerExecutive/Leadership
Level 5 · Reporting & Intelligence Layer
Dashboards & Analytics
PPE compliance rateMonthly / weekly trendsHigh-risk zonesRepeat offenders & recurring risksUnsafe actions by department
AdminHSE ManagerFacility ManagerExecutive/Leadership
Compliance Reporting & Audit Center
Auto-generated compliance reportsTimestamped event logsCamera-level risk scoringExportable audit documentation
HSE ManagerExecutive/Leadership
Resource & Knowledge Hub
Central library for SOPsTraining filesSafety templatesUpdated guidelines
AdminHSE ManagerFacility Manager
Document Center
Exported reportsEvent logsSystem-generated documentsAudit files
AdminHSE ManagerFacility Manager
Design language

A system you can build with.

The palette and type, rebuilt in code; the component kit dropped straight from the Figma library — buttons, inputs, controls, tabs, chips, and tooltips, all on one set of tokens.

Foundations
Indigo#6F53FF
Teal#07D8B9
Blue#1A84D0
Purple#A048D6
Aa
  • Display48 / 600
  • Body16 / 400
  • Label12 / 500
Primary Indigo 500 #6F53FF
Buttons & forms
Figma buttons — Run Scan, Cancel, and icon button across every state
Figma input field with label, helper text, and link Figma input field — typing state Figma input field — error state
Figma radio buttons Figma checkboxes Figma toggles — on and off
Components
Figma chips — selected, avatar, and dismissible
Figma primary tooltip with title, content, and button Figma light tooltip
Figma tabs — active, idle, disabled

From a camera feed to a decision.

The same pipeline runs end to end. A detection becomes an event, an event becomes a record, a record becomes an action assigned to a role — with the heat-map feed, severity, and history all in one view.

AmanVision camera detail — a live industrial feed with a heat-map risk overlay and a side panel of detected events
Real-time monitoring

A live feed that flags risk as it happens.

A heat-map overlay on every camera surfaces PPE gaps, access breaches, and unsafe actions the moment they occur — each one becomes a scored event, not an after-the-fact report.

AmanVision dashboard — total and active alerts, average resolution time, accuracy rate, and an alerts-over-time chart
Decision-ready dashboard

The whole site, at a glance.

Active alerts, average resolution time, and detection accuracy up top; alerts over time and breakdowns by facility and risk scenario below — enough to act on without digging.

AmanVision alerts queue — a table of triggered alerts with time, alert name, risk scenario, location, and status
Incident review

One queue for every event.

Each detection lands with its time, risk scenario, location, and status. The team validates true alarms, flags false positives, and escalates what matters — all in one place.

AmanVision facility detail — a warehouse with its cameras, each showing location, risk scenarios, schedule, and active alerts
Sites & cameras

Configured around the real site.

Facilities, areas, and cameras set up per zone — each with its own risk scenarios and schedule, and a live active-alert count so nothing hides.

A safety platform that fits the region it serves.

AmanVision shipped as a GCC-first HSE platform — one system in place of the spreadsheet-and-paper sprawl, with AI output translated into work every role can trust.

Signals
1

platform in place of spreadsheets, chat threads, and paper across the safety workflow.

5

roles — Admin, HSE Manager, Facility Manager, Safety Officer, Leadership — in one workspace.

Live

risk surfaced the moment it happens, not buried in an incident report written days later.

01

GCC-first by design

Aligned with Saudi regulation and Vision 2030, bilingual, and shaped around local site conditions — not a Western tool in translation.

02

AI made legible

Detections arrive as scored, explained events a non-technical officer can verify and act on in seconds.

03

Audit-ready output

Every event becomes structured, exportable documentation — built for the inspection it’s going to face.

AI is only useful when a person can act on it. The design job was translation.

The model could flag a violation in milliseconds — but that means nothing until an HSE manager trusts it, a safety officer can verify it, and leadership can see the pattern behind it. Most of the work wasn’t the detection; it was turning raw output into something five different roles could read, question, and act on — in their language, against their rules, fast enough to matter.

More case studies