Product Design
Research
Human AI
Designing for the operator who cannot afford to click twice
A conceptual redesign of an emergency command center that compresses the first critical decisions in wildfire response from nearly 48 minutes to 90 seconds.
Platform
Desktop Based
Year
2026
Status
Live
Tools
Figma, Figjam, Framer
At A Glance
A command center for the first ninety seconds
FlytBase Design Fusion S3 asked teams to reimagine the interface for autonomous drone emergency operations. Team KRANTI treated the bottleneck as the operator’s mind in fight-or-flight, not drone speed. I led UX research and every dashboard decision.
Wildfire response is a time story. The traditional first window is close to forty-eight minutes: detect, interpret, decide, dispatch. Autonomy already shaves the flying. It does not shave the staring. Someone still has to look at a wall of feeds and not freeze.
We did not design a prettier operations room. We designed the first ninety seconds of a human who has not slept, with three drones, a heat bloom, and an AI that has already made a recommendation they cannot see.
Challenge
FlytBase Design Fusion S3
Focus
Command center for autonomous drones
Real problem
Human overload, not drone speed
Goal
Read, trust, act in 90 seconds
Deliverables
Figma prototype, FigJam research, Framer site
The Challenge
How might one stressed operator read the situation, trust the AI, and act in 90 seconds?
The user is at 3 AM, with incomplete information, managing multiple drones. Every second of interface confusion is a second of real-world consequence. The brief was not “make this look good.”
Life-critical UI has a different physics. Hover states are a luxury. Nested settings are a liability. A modal that asks “are you sure?” is how a fire gets a second wind. The operator does not want a co-pilot that hides the wheel. They want a co-pilot that points, explains, and gets out of the way.
Constraint
One operator. Multiple agents. Incomplete data. Night shift. No time for a tutorial.
Job
Compress detect → interpret → decide → dispatch into a glance, a why, and one confirm.
Failure
Equal-weight alerts, invisible AI, and menus designed for a rested analyst.
Role
UX research lead and owner of dashboard decisions for Team KRANTI, from insight to high-fidelity.
Problem
The bottleneck was never the aircraft
Traditional response to a detected wildfire takes close to 48 minutes. Operators cannot tell what the AI already decided, alerts arrive at equal volume, and existing tools were designed for calm, not crisis.
“I don't need the system to decide for me. I need it to tell me why it thinks that.”
UAV operator, interview
That sentence became the product. Autonomy without explanation is a black box. A black box, under stress, is ignored or worse, obeyed without understanding. Both are accidents waiting for a timestamp.
We also heard a quieter fear: not “where is the fire,” but “is someone else already going.” Coordination debt looks like hesitation. Hesitation looks like incompetence. It is usually a missing column in the UI.
Cannot tell what the AI decided versus what needs a human.
Too many alerts erode trust.
Interfaces built for calm, not crisis.
Under pressure, people drop analysis and use instinct.
Hesitation when unsure others are already responding.
People
Two operators, one glance
We designed for an experienced swarm operator and a newer responder who still needs guidance. Both are working alone, at night, with incomplete information.
The experienced operator does not want a wizard. The new responder cannot survive without one. The same layout had to serve both: hierarchy so the veteran can skip, explanation so the new person does not have to open a second manual. If the UI had two modes, it would fail at 3 AM when nobody remembers which mode they are in.
Primary
Experienced operator managing multiple drone swarms, Needs one-glance situation picture, not menus
Secondary
Newer responder who needs guidance without overload, working at 3 AM with incomplete information
Approach
Research before a single pixel
Secondary research into emergency protocols, Kahneman’s System 1/2, and human–AI trust. Personas and journeys were drawn only for the first 90 seconds. Every idea was filtered by one question: does this reduce cognitive load?
We started with human factors, not with a component library. Under time pressure, people abandon System 2. An interface that requires comparison across four panels will be abandoned. So we designed for instinct: one critical object, one recommended action, one override.
Trust research was equally practical. More telemetry does not create trust. Reversible, explained recommendations do. That is why “why” sits next to “confirm,” and why override is a peer of confirm, not a nested overflow item.
01
Alert
Fire detected
02
Read
One-glance picture
03
Trust
Sees the AI’s why
04
Act
Confirm or override
05
Monitor
Drones coordinate
Secondary research: protocols, psychology, human–AI trust
Personas and journey mapping for the first 90 seconds
First-principles ideation to cut cognitive load
Dashboard architecture around confidence and confirmation
High-fidelity design: clear, explainable, easy override
Research Insights: Insights sorted from research into observations, pain points and ideas.
User Flow: The path a user takes through the product, including the decisions along the way.
Design & Solution
When seconds save lives, the interface cannot be the bottleneck
The UI is built around confidence scores and confirmation actions rather than raw feeds. Critical information is visible without menus. Override is one action, not a nested dialog.
The map is not decoration. It is the situation. The right rail is not a chat. It is the recommendation with a spine: what, why, confidence, confirm, override. Before-and-after was the argument to judges: equal-weight lists are how operators drown; one critical card is how they swim.
Confidence scores
Every AI suggestion shows how sure it is, and why.
One-glance awareness
Critical information is visible without opening a menu.
Effortless override
The human can take control in one clear action.
Less alert noise
A strict split between critical and informational items.
Transparent coordination
See what each drone and agent is already doing.
Dark, high-contrast field
Designed for a 3 AM operations room, not a marketing site.
Principals
Five filters for every layout
If a control needed a tooltip to be safe, it was not safe. If a recommendation could not show its work, it could not be the default. Those rules deleted more frames than they created.
Remove noise
/
So what matters is instantly visible.
Support fast decisions
/
While keeping human override effortless.
Explainable AI
/
Every recommendation is transparent and reversible.
Multi-agent coordination
/
Without adding cognitive load to the operator.
Share the pressure
/
The system works alongside the human, not against them.
Information architecture: How the product is organised, from entry points down to the deepest screens.
Decisions
A complete challenge package, not a moodboard
This is a conceptual redesign. We did not put drones in a fire. What we did ship is a research-backed prototype, a FigJam of the argument, and a presentation that can be judged. The next honest step is a stress test with operators, not another visual pass.
Designed to lower heart rate
Serious, calm under pressure, trustworthy. Dark command-center surfaces, high contrast, minimal noise, and a tone that is direct and respectful of the operator’s expertise.
Outcome
From pain point to design decision
Dark UI is often aesthetic. Here it is environmental: a lit white dashboard in a dark room is a second emergency. We still kept cream paper for this case study so the work can be read in a portfolio; the product itself stays night-native.
What works
Could improve
Next
What this project taught
Life-critical seconds change the brief. Normal dashboard patterns break.
Great work is often deciding what the operator never has to think about.
Trust comes from explainability and easy override, not more data.
The interface should support fast instinct while the human stays in control.
FAQ
Questions I get asked a lot
Anything else, just send me a note, I answer within a working day.