Loading Experience0%
A command centre for autonomous drone security operations, designed for a design challenge from FlytBase, set at Le Musée d'Art Précieux in Paris.
Role
Product Designer
Team
Solo
Timeline
2 weeks
Tools
Figma, Stitch, Claude Code
I designed a command centre for a security team managing a fleet of autonomous drones at a museum. FlytBase set this as a design challenge, and I kept building on it afterward.
Security operators face information overload from disconnected systems: CCTV, motion sensors, pressure sensors, and drone consoles. There is no single source of truth.
"With 65,000 square metres and over 400 rooms, it's impossible to have eyes everywhere. We almost lost a Degas worth €30 million when sensors failed in the east wing."
Marc
Head of Security, Le Musée d'Art Précieux
Disconnected CCTV, motion sensors, and access control systems create information silos. This makes it hard for operators to build a clear picture during an active incident.
150+
Global 2023
Attempted thefts
€30M
At Risk
Asset Value
45s
Lost Time
Manual Triage
24/7
Requirement
Operational
I mapped four people who use this system, based on the roles in the brief.
Security Director (System role: Director)
Wants accountability and clear, trustworthy performance data. Avoids visual clutter.
Lead Drone Operator (System role: Team Lead)
Calm under pressure. Needs quick triage support and a clear flow when more than one incident happens at once.
Junior Security Analyst (System role: Field Operator)
Comfortable with technology but still learning the job. Needs a confidence signal on each alert so he escalates the right ones.
Night Shift Supervisor (System role: Admin)
Coordinates field guards and manages system settings. Needs full context at the start of each shift.
I built the dashboard around a fixed set of rules. Here is how it decides what happens next.
Example recommendation
Assign Sentinel-2 to North Storage Wing (gap 45 min). Sentinel-2: Idle, 100% battery, health 96. Watcher-3 skipped: health 75, flag: Lens calibration requested.
Shift Handover
Before a shift starts, the operator goes through a mandatory briefing. It has 5 modules, so the next person on duty starts with full context instead of a blank dashboard.
Shift briefing screen, with the 5 modules listed above.
I'm showing 6 screens here. The live prototype has the rest.
The 2D museum floor plan, with live alerts and SLA countdowns.
Drone health and status for the whole fleet.
Manual override, with the 5 minute cap and warning states.
Waypoints, blind spots, and the rule-based patrol recommendation.
The incident database.
Roles and permissions.
All major user flows are mapped across the Dashboard as the central hub, from alert triage to shift handover, and from drone dispatch to manual override. The flowchart has 12 flow groups, including alert management, drone patrol, manual control, multi-incident mode, shift briefing, fleet management, settings, and escalation paths.
This flowchart is the source of truth for the prototype.
I built the first version of this prototype in one evening and night, starting on 25 December 2025. It took 42 prompts. I hit syntax errors 3 times and a blank deploy 4 times before it worked. Stitch generated the screens. The model suggested names for some museum wings, and I replaced them with my own.
Perplexity
Deep analysis of the brief: extracted museum context, constraints, and all 4 persona pain points
Perplexity + ChatGPT
Problem framing, grouped issues into 3 gaps: decision support, multi-incident prioritisation, context persistence
ChatGPT + Perplexity
Information architecture: wrote textual user flow descriptions before any visual structure
Eraser.io → FigJam
Flowchart: translated text flows into structured nodes, then iterated and annotated in FigJam
ChatGPT
Flow validation: checked every step against each persona, which led to making the shift briefing mandatory
AI + Gemini Dynamic View
Feature lists per screen; visual layout exploration for alert cards, multi-incident view, shift briefing
Stitch + Google AI Studio
Mockup refinement for UI consistency; built interactive prototype simulating a 2:37-2:47 AM scenario
Perplexity + ChatGPT
Prototype stress-testing: verified every pain point was addressed, and found gaps in drone readiness cues
I used Claude Code for a later rebuild of parts of this prototype.
Every decision was a trade-off. I chose operational safety over decorative features.
Decision 01
Clarity over complexity
Instead of: Decorative data visualisations
A calm, clear layout works better than detailed charts at 2 AM during a live incident.
Decision 02
Human in the loop
Instead of: Full automatic dispatch
A person makes the final call on dispatch, not a model.
Decision 03
A drone health gate
Instead of: Letting an operator assign any drone
A drone with a fault or low battery should not be sent on a task. I added a check: status must be Idle or Active, and battery above 50%.
Decision 04
A 5 minute cap on Manual Control
Instead of: An open-ended manual session
A timed session with warnings at 2:00 and 4:00 keeps an operator from losing track of time while piloting one drone by hand.
I built this to WCAG 2.2 AA targets, and checked it with axe-core, a keyboard pass, and a contrast script. Known limits are listed in the project README. I have not tested it with a real screen reader.
This prototype is a working demonstration of the rules above. Here is what it actually does.
Previous Project
Samsung PRISMNext Project
GuruVR Metaversity