"It feels like the whole day is on hold."
Fixie AI
Timeline
Dec 2025 - Feb 2026
TEAM
Individual
MY ROLE
UX Strategy
UI & Interaction
AI Workflow
Rapid Prototyping
User Research
In most repair flows, you report the problem and then you're left guessing. No update, no timeline. Fixie AI works in the gap between the report and the fix is where a stalled repair costs a home a day, and a restaurant its business.
GYUHO Lee © 2026
Interact
Users don't just want a fast fix. They want a transparent process.
01. Highlights
No guessing.
No waiting.
Just what's next.
AI Diagnosis
Vague symptoms, clear data
Before, a broken machine meant guessing, then waiting for a callback. Fixie AI reads the symptoms you can describe and turns them into a diagnosis a technician can act on.
Dynamic Matching
Watch the repair unfold
Each step shows up as it happens, so no one has to call and ask "where are we now?"
Fast Response
Urgent jobs, surfaced first
Technicians see urgent jobs first, send a first estimate from the details Fixie AI gathered, and head out once the customer approves the estimate.
In prototype testing, the matched flow brought the first estimate from 11–14 days to under 30 minutes.
02. Process
Human judgment. AI speed.
Challenge
AI fills the gaps with averages
AI is fast, but it reaches for the average answer. In a repair flow shaped by anxious users and field constraints, the average answer quietly erases the details that matter. Speed without judgment just produces confident, generic output faster.
Solution
Decide where to let go, and where not to
I mapped the points where a wrong default would cost the most, and kept human judgment there. AI runs the rest. The skill wasn't using AI or avoiding it, but knowing exactly when to step in.
Macro Framework
The principle, in three moves:
Phase 1.
Brain (Human)
Role - Empathy, Definition
Output - Problem Statement
Phase 2.
Spark (AI)
Role - Ideation, Variations
Output - Raw Layouts
Phase 3.
Hand (Human)
Role - Logic Check, Edge Cases
Output - Final Polish
4 Step AI–Human Collaboration
Below, the same principle applied step by step in this project. Every screen was tuned around three signals: clarity, urgency, and trust.
Step 1.
Research & Synthesis
Human: Interviewed both sides of the repair, the owners who report a breakdown and the technicians who fix it.
Sample: 14 people who had been through an equipment repair informed the project. The 7 closest to the breakdown, the restaurant owners and the technicians, gave in-depth interviews that drove the core insights.
AI: Clustered scattered complaints and reviews into candidate pain points.
Validation (human): I checked every cluster against the original interviews and re-grouped by hand, so each pain point traced back to a real account, not a keyword match.
Result: A pain-point map traceable to specific interviews, not assumptions.
Step 2.
Information Architecture
Human: Set what information comes first, balancing the user's task flow with what keeps them engaged.
AI: Used AI to stress-test the flow for gaps and edge cases I'd missed.
Result: An information structure that holds up when the user is anxious and rushed.
Step 3.
Rapid Visual Exploration
AI: AI generated low-fidelity UI options in minutes.
Human: I reframed the structures the AI couldn't see, the field constraints and the emotional context.
Result: Less time drawing screens, more time on the decisions that mattered.
Step 4.
Prototyping & Refinement
Human: Designed the technician screen around urgency: a visible time-limit bar and a bonus for fast acceptance, turning a plain UI into a behavioral nudge.
AI: Used AI to check interaction timing, flag overlooked states, and suggest smoother interaction patterns.
Result: A high-fidelity prototype where urgency is designed in, not left to chance.
03. Problem Space
Pending. No ETA.
In most repair systems, you report a problem and then go quiet, no update, no timeline. Working at a restaurant in Berlin, I watched a minor repair stall for days because no one could say what came next.
Fixie AI works on both sides of that gap: the person who needs the fix, and the technician who takes it on.
UX Breakdown
Key Friction Points
Where it broke down
Core Insight
Key Friction Points
Communication gaps
After reporting an issue, people had no idea what came next, how long it would take, or whether anyone understood the problem.
"I wasn't sure what was happening, but I had to pretend I understood."
No one could say when
No reliable signal of when a technician would arrive, or when the problem would actually be solved.
"I keep checking my phone. It feels like the whole day is on hold."
Silence after the first report
Once the report was in, progress disappeared from view, leaving people to chase updates themselves.
"If one small part is missing, I lose another day of business."
Key Friction Points
The black box
You might know how repairs work in theory.
You still can't see where yours actually is.
Quick Insight
Technicians were willing.
The blockers were manual work and missing data.
3 technicians · 4 shop owners · Berlin
The Anatomy of Friction
The frustration was real. Its cause was structural.
Mapping the journey surfaced two points where the system went silent.
Peak 1. Booked, then silence
A confirmation gap. The only confirmation was a receipt. After that, people had to chase every schedule change themselves.
Peak 2. Arriving without the part
A prediction failure. With no live inventory and no on-site briefing, technicians often arrived without the right part. A few repeat visits cost a full day of downtime. All of them cost the owner time or money.
The questions
left unanswered
“When are they actually coming?”
“How long is this going to take?”
“Will it be fixed properly?”
“Do I need to keep following up?”
Representative quotes from customer interviews.
The Core Insight
From 'Fragmented Steps' to a 'Synchronized Loop'
When people reach for AI or call around for answers, it's rarely impatience. It's uncertainty. The real problem was never the speed of the repair, it was the silence around it.
Strategic Definition
So the solution stopped being a matching system. It became a communication layer that removes the black box between diagnosis and repair.
Design Principle
'Zero Guesswork'
People should never have to guess what's happening. I cut visual noise and pushed the key information, status, time, steps, to the front. The point wasn't smarter AI. It was clearer information that lets people act with confidence.
Integrated Strategy Map
Where friction meets the fix
Each user-facing friction, traced to its business root and matched to a design response.
Challenge
to Strategy
1.
Operational Inefficiency
Targeting a "Zero-Guesswork" Experience
Challenge: Information Asymmetry
Technicians often arrive without enough information about the issue.
Strategy: AI Pre-Diagnosis
Converts user descriptions into clear, structured data so technicians are prepared before arrival.
Goal: Fewer, better-prepared visits.
Replaces blind visits with better-prepared service.
2.
Market Bottleneck
Improving Match Speed Through Incentives
Challenge: Supply-Demand Imbalance
Matching delays during peak hours or for urgent tasks, leading to lost opportunities.
Strategy: Dynamic Time-Deal
Highlights high-value jobs in real time to encourage faster responses.
Goal: Faster matches when it's urgent
Ensures urgent needs are met more quickly.
3.
Process Uncertainty
Building Trust by Minimizing "Anxiety Costs"
Challenge: High Anxiety Costs
Users feel stressed due to a lack of visibility during the repair process.
Strategy: Real-time Status Tracker
Provides a clear, trackable timeline that shows progress and builds confidence.
Goal: Build a Trust Loop
Encourages repeat use by making the process transparent and predictable.
04. Strategy
Three pillars. One goal
: end the uncertainty.
AI doesn't replace technicians; it replaces uncertainty.
Why AI?
People already lean on AI to make sense of everyday problems. So I placed it exactly where the anxiety peaks, in the gap between diagnosis and action. It doesn't do the expert's job. It structures the messy, incomplete signals on both sides, so the user and the technician finally work from the same picture.
It structures incomplete signals so users and technicians can operate with a shared understanding.
Step 1.
The Problem (Uncertainty)
User: anxious, guessing at what's actually wrong
Technician: heading out half-blind, no real data about the site.
Step 2.
AI Intervention (Structuring Signals)
AI turns scattered input into a structured brief, syncs the timeline, and gives the technician a real picture before arrival.
Step 3.
Human Action (Confident Execution)
User: calm, because the process is finally visible.
Technician: arriving prepared, with the right parts and tools.
The logic the product runs on, from the user's first tap to the technician's arrival.
3 UX Design Pillars
Three design pillars, one for each friction the product had to solve.
01.
The Translator
AI Data Structuring
Turns a user's rough description into a structured brief the technician can read at a glance.
The technician sees the likely issue before leaving, so fewer trips end in "I'll have to come back."
Prepared visits
02.
The Matchmaker
Dynamic Incentives
Raises the reward on urgent jobs in real time, so the ones that matter get picked up first.
During peak hours, urgent tasks surface first instead of sitting in a queue.
No job left waiting
03.
The Zen Master
Timeline Clarity
A live timeline that shows exactly where the repair stands, no need to ask.
When people can see the whole process, they come back, and they recommend it.
Trust that earns the next call
05. Exploration
AI explores. I decide.
Design Goal
To move fast on early exploration with Google Stitch, so my own time went to the structural calls AI couldn't make.
AI Exploration,
Human Judgment
Google Stitch
Google Stitch generated a wide range of IA layouts in a fraction of the usual time. I compared them and kept the structure that read most intuitively.
My Role
I reset the visual priorities around how the service actually works, and reworked the AI drafts so the data held up in real-world use, not just on screen.
→ What follows isn't the final product, it's the decisions that shaped it. Where the AI draft fell short, and what I changed.
The Translator
One structure, both sides
Problem
The Translator turns a user's plain-language description into a structured diagnostic brief, the part you've already seen. The same structuring had to work on the technician's side too, where the call was how to organize those briefs at scale.
AI Assistant
Fixie
Analyze and diagnose the issue with specialized technical AI
Connect users with professional technicians when needed
Before matching users with a technician, the system proposes self-repair options.
This prioritizes user autonomy and potential cost savings.
How does the technician see this?
The diagnostic brief gives the technician the error codes and likely parts up front, so the first visit is the one that fixes it.
Why job codes, not names
The AI draft sorted jobs by customer name. That breaks the moment a technician juggles ten jobs at once.
A consumer app can lead with names or locations. A technician's tool needs a system that scales, indexing that holds up across dozens of active jobs.
So I switched to unique job codes (Job #4092), so a technician can find and track any case in seconds.
The Matchmaker
Matching that works
in the field
Independent technicians often rely on manual workflows and unstable customer acquisition.
To address this, I introduced a ride-hailing–inspired model, integrating automated matching and structured workflows directly into the dashboard.
Two modes, Standard and Urgent, structure workflow around Urgency.
Decision order: Urgency → Value → Fit
Urgency Drives Earnings
Choosing "ASAP" raises the technician's payout, so urgent jobs get picked up first instead of sitting idle.
The reward is visible before the technician accepts, a lower platform fee on urgent jobs, so the incentive is obvious, not buried.
Technicians - sees the higher payout up front.
Customers - gets matched faster.
Deep Dive: Resolving UX Conflicts
While prototyping the task management tab, the AI-generated UI surfaced interaction conflicts that static mockups had hidden.
Interaction Conflicts
Map vs. Card Gestures
Overlapping scroll areas made one-handed interactions unreliable.
Reduced Map Visibility
Large cards reduced map visibility, making routes harder to scan.
The interface needed to adapt to real field behavior, not just visual layout.
View AI Draft → Final UI
Separated Interaction Zones
Horizontal scrolling separated task browsing from map interaction, reducing gesture conflicts.
Adaptive Viewport
Once navigation starts, the task card minimizes to keep the route visible while remaining easy to revisit.
The Zen Master
Operational Transparency
A shared timeline put every step in view, the same picture for the user and the technician at once.
Shared Visibility
The user's progress view and the technician's actions update from the same source, so neither side is guessing.
Reduced Workflow
The technician's updates collapse into four large taps, fast enough to use mid-repair, hands full.
PING System
The restaurant owner flagged a recurring complaint: customers couldn't get an update without calling, again and again.
"Can technician status be checked in just a few taps?"
Side Effects :
More options ➔ More hesitation ➔ More confusion
So I cut it down with the restaurant owner, to the three questions customers actually asked. Send, and wait, no calling.
Response Windows & Accountability
The 10-minute window isn't a deadline, it's an offer. Respond in time and the urgency bonus is yours. Let it pass and the request stays active, just without the extra.
Respond in time and the +€8 urgency bonus from the matching step is yours.
Repeated misses lower the technician's reliability score, so fewer jobs surface next time.
Motion as a signal
Two moments use motion on purpose.
I tested both versions with potential users. Faster, brighter motion mostly read as an alert or error, while the slower interaction was the one people called calming.
Slow pulse — the active step breathes, signaling calm.
It's moving, you don't need to do anything.
Live countdown — the response bar shrinks, making urgency physical. Now matters.
"The pulsing motion settled the tension."
— Minjun, works at an Asian restaurant
The interface speaks before the text does.
Digital Handshake
To close the loop with trust, not a surprise bill, the system adds one confirmation step between technician and customer.
In Germany, the invoice often lands days after the work, with no warning, the final cost is a surprise. The process just... ends, without a clear sense of closure.
Interviews surfaced a simple fact: a technician can give a rough estimate the moment they spot the issue. That was the opening, align on cost early, before the work, not after.
Once the repair is complete, the technician finalizes the work and submits:
AI-summarized repair report
Final invoice
Early alignment up front, clean confirmation at the end, so the last thing the customer feels is closure, not a surprise.
Supporting Interactions
Role-based entry
Users and technicians land in different flows from the first screen.
One place for every repair
Active jobs and full history, side by side.
See full breakdown
Pay with the full picture
Final cost, the work done, and proof, all before payment.
06. Impact
Before and After.
In Numbers.
This project is still at MVP stage with no live metrics yet.
Manual booking is measured from 14 past customer requests. Matched flow time comes from a prototype test with 7 technicians, directional given the sample size.
Time to first estimate
Manual booking
11–14 days
14 customers, past requests
Matched flow
under 30 min
7 technicians, directional test
Same task, different processes: getting to a first estimate. Manual booking took 11-14 days for customers. In a small technician test, the matched flow got that down to under 30 minutes.
Diagnostic loop
Structured AI intake replaces the phone-and-email ping-pong. The diagnostic loop shrinks from 5–7 days of back-and-forth to near real-time.
Fastest gains
UI exploration
AI-generated drafts cut the time spent laying out screens.
Logic validation
Edge cases showed up before they reached the prototype, so I could move faster.
Research synthesis
Stacked interview notes and needs clustered into clear patterns in minutes.
Where AI saved me the most time
This one isn't about the user, it's about my own process. Using AI for the slow early stages cut mock-up production from 3–4 weeks to about 1.5, freeing the time for the structural calls that needed me.
Small sample, early stage, but across all 14 the direction held
: less waiting, less guessing, less friction.
"
Beyond individual repairs
Fixie closed three separate delays: matching, pre-arrival information, and visibility during the repair itself. Most of that delay ran in both directions. Requesters and technicians were often waiting on the same gap, just from opposite sides.
What Mattered Most
Field constraints shaped this more than any visual preference. The UI followed thumb-zone reach, one-handed use, and real movement, reflecting how technicians actually work rather than how they were supposed to.
AI accelerated the process, but judgment determined its direction. The real skill was knowing when to override, when to reframe, and when the output simply wasn't good enough.
"NEXT ARCHIVE"
nooding
DWELL
