Go back

Say “Hello!”

Go back

Say “Hello!”

Go back

Say “Hello!”

Company

GeoMate

Year

2025

Scope of work

Web app re-design

User flow re-design

Main stack

Industry

Mapping

AI

Autonomous Driving

Simulation

Company

GeoMate

Year

2025

Scope of work

Web app re-design

User flow re-design

Main stack

Industry

Mapping

AI

Autonomous Driving

Simulation

Redesigning Simulation Mapping for Iterative Workflows

The Challenge: RealSimE's MVP couldn't support real production workflows—users lost work between sessions and faced opaque export processes. With October client trials approaching, the platform risked failing to demonstrate value in the competitive autonomous driving simulation market.


The Impact: Transformed RealSimE from proof-of-concept into production-ready platform—achieving 95% task completion in beta testing, reducing workflow time by 58%, and establishing foundational patterns that now power GeoMate's entire product ecosystem.

View live product

The impact

60% Faster

60%
Faster

Task time: 7.5 min → 3 min, reducing 40% abandonment rate

Task time: 7.5 min → 3 min, reducing 40% abandonment

100% Persistence

All 25 beta users saved/resumed regions vs 0% in MVP

All 25 beta users resumed regions vs 0% in MVP

85% Feature adoption

New features actively used vs 30% in MVP (n=25)

Zero token disputes

Export transparency eliminated unexpected consumption complaints

95% less confusion

"How do I start?" dropped from 60% → 5% via onboarding iteration

60% Faster

Task time: 7.5 min → 3 min, reducing 40% abandonment rate

100% Persistence

All 25 beta users saved/resumed regions vs 0% in MVP

85% Feature adoption

New features actively used vs 30% in MVP (n=25)

Zero token disputes

Export transparency eliminated unexpected consumption complaints

95% less confusion

"How do I start?" dropped from 60% → 5% via onboarding iteration

Why it mattered

RealSimE operates in the autonomous driving simulation market, where engineers need high-fidelity HD maps to test ADAS systems in virtual environments like CARLA, dSPACE, and RoadRunner. Unlike consumer mapping tools, simulation workflows require:


  • Iterative testing: Engineers test hundreds of scenarios across multiple regions

  • Semantic precision: Lane topology, traffic signals, intersection geometry

  • Format flexibility: OpenDRIVE, Lanelet2, USD, GeoJSON export


With October 2025 client trials 12 weeks away, RealSimE’s MVP was blocking adoption—threatening commercial launch viability.

Why it mattered

RealSimE operates in the autonomous driving simulation market, where engineers need high-fidelity HD maps to test ADAS systems in virtual environments like CARLA, dSPACE, and RoadRunner. Unlike consumer mapping tools, simulation workflows require:


  • Iterative testing: Engineers test hundreds of scenarios across multiple regions

  • Semantic precision: Lane topology, traffic signals, intersection geometry

  • Format flexibility: OpenDRIVE, Lanelet2, USD, GeoJSON export


With October 2025 client trials 12 weeks away, RealSimE’s MVP was blocking adoption—threatening commercial launch viability.

The Problem (Quantified)

The Problem

Quantified pain points (n=12 internal testers)

  • Zero work persistence
  • 7.5 min average task time
  • 40% abandonment rate
  • Rectangle-only limitation
  • No export transparency
  • Visual clutter (Mapbox defaults)
  • Visual clutter

The Problem (Quantified)

Quantified pain points (n=12 internal testers)

  • Zero work persistence
  • 7.5 min average task time
  • 40% abandonment rate
  • Rectangle-only limitation
  • No export transparency
  • Visual clutter (Mapbox defaults)

Real user feedback

“I spent 15 minutes selecting regions yesterday, logged in today, and everything was gone.”


“I can’t test multiple intersections without losing track.”


“I used 2000 tokens on building data I didn’t need—there was no warning.”

Strategic Foundation

01

02

Understanding the workflow gap

Workflow gap

Through user interviews (n=8) and internal alpha testing (n=12), I identified a critical pattern:


AD engineers test iteratively: Selecting multiple regions, comparing intersection configurations, and refining scenarios over weeks. RealSimE's MVP had zero spatial memory. Regions disappeared after export, forcing users to manually re-find locations every session.


The core insight:We didn't need better drawing tools first—we needed persistent region management. Enabling users to save, name, and revisit regions would unlock iterative work.

Project constraints and prioritization

Project constraints

Constraint: 12 weeks to October client trials, with 17 feature requests from internal team and customer feedback—but limited developer capacity meant not everything could be built in time.


Facilitation: I led a prioritization workshop with founders, product lead, and development team, using an Impact vs Effort analysis to sequence delivery effectively.

Key trade off

High-effort editing features, such as polygon re-editing, were intentionally deferred to focus on foundational persistence and transparency—core capabilities that unblocked critical workflows for trial users and ensured delivery within the 12-week timeline.

Design Solutions

02

1. Flexible drawing tools

Initial problem:

Rectangle-only selection forced inefficient workflows. Users made 3x manual adjustments to capture specific street configurations, wasting time and tokens on unnecessary coverage areas.

Solution: Polygon + rectangle dual tools

  • Toolbar repositioning: Moved controls from bottom-right "Draw boundary" button to top bar alongside search—following mapping tool conventions where primary actions live in headers. Improved discoverability in testing.


  • Polygon precision: Engineers now trace exact road segments—critical for scenarios like intersection or highway merge zone.


  • Real-time feedback: Card auto-updates road length and token cost as users draw. Knowing resource impact during selection (not after export) encourages iterative refinement: "Is this extra block worth 200 tokens?"

1. Flexible drawing tools

Initial problem:

Rectangle-only selection forced inefficient workflows. Users made 3x manual adjustments to capture specific street configurations, wasting time and tokens on unnecessary coverage areas.

Solution: Polygon + rectangle dual tools

  • Toolbar repositioning: Moved controls from bottom-right "Draw boundary" button to top bar alongside search—following mapping tool conventions where primary actions live in headers. Improved discoverability in testing.


  • Polygon precision: Engineers now trace exact road segments—critical for scenarios like intersection or highway merge zone.


  • Real-time feedback: Card auto-updates road length and token cost as users draw. Knowing resource impact during selection (not after export) encourages iterative refinement: "Is this extra block worth 200 tokens?"

2. Persistent region management

2. Persistent management

Initial problem:

Regions disappeared after export—users had no way to save, name, or revisit selected areas. This forced manual re-searching and re-drawing every time they wanted to test a different configuration or compare multiple scenarios.

Solution: Region sidebar with persistent card

  • Card hierarchy：Region name (user-editable) sits at top, with road length and token cost inline below. Surfacing resource impact before export enabled budget-conscious decisions and prevented MVP's token disputes.


  • Inline actions: Three-dot menu embeds Export, Rename, Zoom, Delete directly in cards. MVP forced users into a separate list view, breaking spatial context. This reduced clicks-to-export from 4 to 2.


  • Show/hide toggle: Checkbox declutters view without deleting work. Beta users managing 5-6 areas wanted to temporarily hide regions during complex comparisons.

2. Persistent region management

Initial problem:

Regions disappeared after export—users had no way to save, name, or revisit selected areas. This forced manual re-searching and re-drawing every time they wanted to test a different configuration or compare multiple scenarios.

Solution: Region sidebar with persistent card

  • Card hierarchy：Region name (user-editable) sits at top, with road length and token cost inline below. Surfacing resource impact before export enabled budget-conscious decisions and prevented MVP's token disputes.


  • Inline actions: Three-dot menu embeds Export, Rename, Zoom, Delete directly in cards. MVP forced users into a separate list view, breaking spatial context. This reduced clicks-to-export from 4 to 2.


  • Show/hide toggle: Checkbox declutters view without deleting work. Beta users managing 5-6 areas wanted to temporarily hide regions during complex comparisons.

3.Export workflow redesign

3.Export redesign

Initial problem:

Users exported data blindly with no cost preview or control over data composition. Client feedback revealed unexpected token consumption on unnecessary building data—triggering billing disputes and support tickets.

Solution: three stage workflow

  • Stage 1 - Preview modal: Displays region, format, and token cost calculator (total/used/estimated/remaining). Building data toggle (default OFF) lets users exclude unnecessary geometry. Confirms resource commitment before export.


  • Stage 2 - Non blocking progress: Top banner shows status without freezing map. Users continue drawing other regions—enabling parallel workflows.


  • Stage 3 - Completion: Success offers Download + View OpenDRIVE. Failure shows retry + support contact.

3.Export workflow redesign

Initial problem:

Users exported data blindly with no cost preview or control over data composition. Client feedback revealed unexpected token consumption on unnecessary building data—triggering billing disputes and support tickets.

Solution: three stage workflow

  • Stage 1 - Preview modal: Displays region, format, and token cost calculator (total/used/estimated/remaining). Building data toggle (default OFF) lets users exclude unnecessary geometry. Confirms resource commitment before export.


  • Stage 2 - Non blocking progress: Top banner shows status without freezing map. Users continue drawing other regions—enabling parallel workflows.


  • Stage 3 - Completion: Success offers Download + View OpenDRIVE. Failure shows retry + support contact.

  1. OpenDRIVE Live Preview

  1. Live Preview

Initial problem:

After spending tokens to export, users had to download files and open them in external tools (CARLA, dSPACE) to view OpenDRIVE output. This broken workflow meant leaving the platform just to validate data quality.

Solution: Embedded Split-Screen Viewer

  • Draggable divider: Users adjust split ratio, split-screen maintained spatial alignment for side-by-side validation.


  • Why it matters: Users spot issues (missing lanes, incorrect topology) immediately without downloading files or switching tools—enabling rapid iteration.

  1. OpenDRIVE Live Preview

Initial problem:

After spending tokens to export, users had to download files and open them in external tools (CARLA, dSPACE) to view OpenDRIVE output. This broken workflow meant leaving the platform just to validate data quality.

Solution: Embedded Split-Screen Viewer

  • Draggable divider: Users adjust split ratio, split-screen maintained spatial alignment for side-by-side validation.


  • Why it matters: Users spot issues (missing lanes, incorrect topology) immediately without downloading files or switching tools—enabling rapid iteration.

Iteration & Refinement

02

Maze usability testing

After completing Version 1.0, I conducted Maze usability testing with 25 internal beta users to validate the redesigned workflow. Testing revealed two critical friction points requiring rapid iteration:

Post-launch onboarding refinement

Onboarding refinement

Testing result

60% of users asked "What do I do first?" after landing on the map.


Task: "Select a region and export OpenDRIVE data." Average time to first successful action: 2.1 minutes—too high for a streamlined tool.

Solution: Monochrome Semantic Map

  • Iteration 1 - Hover Tooltips: Added explanatory text on icon hover.


  • Iteration 2 - Welcome Modal: Large modal on first login with "Get Started" CTA.

Post-launch onboarding refinement

Testing result

60% of users asked "What do I do first?" after landing on the map.


Task: "Select a region and export OpenDRIVE data." Average time to first successful action: 2.1 minutes—too high for a streamlined tool.

Solution: Monochrome Semantic Map

  • Iteration 1 - Hover Tooltips: Added explanatory text on icon hover.


  • Iteration 2 - Welcome Modal: Large modal on first login with "Get Started" CTA.

Quick download

Quick download

Testing result

"Why do I have to open sidebar → find card → click three-dots → export? I just want to download this region I'm looking at." Users wanted faster access to export actions without leaving map context.

Solution: Monochrome Semantic Map

  • Right-click activation: Users right-click any region polygon on map to trigger context menu. Familiar pattern from design tools and file managers—low learning curve for technical users.


  • Inline region info: Menu displays region name, road length, and estimated token cost—same metadata as sidebar cards, but without navigation.


  • One-click export: "Export Data" button triggers the same preview modal as sidebar workflow, maintaining consistency while reducing steps.

Quick download

Testing result

"Why do I have to open sidebar → find card → click three-dots → export? I just want to download this region I'm looking at." Users wanted faster access to export actions without leaving map context.

Solution: Monochrome Semantic Map

  • Right-click activation: Users right-click any region polygon on map to trigger context menu. Familiar pattern from design tools and file managers—low learning curve for technical users.


  • Inline region info: Menu displays region name, road length, and estimated token cost—same metadata as sidebar cards, but without navigation.


  • One-click export: "Export Data" button triggers the same preview modal as sidebar workflow, maintaining consistency while reducing steps.

Key learnings

Prioritization & fast shipping

Impact/Effort matrix turned 17 features into sequenced roadmap. Deferring polygon re-editing enabled 12-week delivery—shipping fast > chasing perfection for early validation.

Transparency reduces risk

Token preview + building toggle cut rejection 40% → 8%. Split-screen validation (85% adoption) caught issues pre-download. Visibility > guesswork.

Iterate until clarity emerges

Onboarding evolution (Tooltips → Modal → GIF) reduced confusion 60% → 5%. First impressions determine if users explore or bounce— test ruthlessly.

Domain expertise > polish

Monochrome maps serve AD engineers' semantic needs (lane topology, intersections) over aesthetics. Collaborate with domain experts to prioritize correct data.

Prioritization & fast shipping

Impact/Effort matrix turned 17 features into sequenced roadmap. Deferring polygon re-editing enabled 12-week delivery—shipping fast > chasing perfection for early validation.

Transparency reduces risk

Token preview + building toggle cut rejection 40% → 8%. Split-screen validation (85% adoption) caught issues pre-download. Visibility > guesswork.

Iterate until clarity emerges

Onboarding evolution (Tooltips → Modal → GIF) reduced confusion 60% → 5%. First impressions determine if users explore or bounce— test ruthlessly.

Domain expertise > polish

Monochrome maps serve AD engineers' semantic needs (lane topology, intersections) over aesthetics. Collaborate with domain experts to prioritize correct data.

Next project

Connecting 800,000+ Users
Across 130+ Countries

Read full story

Next project

Connecting 800,000+ Users
Across 130+ Countries

Read full story

Works

About

Linkedin

Email

X

© Tina Tan 2025

Works

About

Linkedin

Email

X

© Tina Tan 2025

Works

About

Linkedin

Email

X

© Tina Tan 2025