Back to portfolio

3D Building Tools

A tool I helped build in 2022 had outgrown itself. I rebuilt it into one system — Simple Map View — then handed it to engineering to run.

Role
Design & build the rebuild · AI-assisted
Problem
Aging tool, fragmented maps, no room to grow
Outcome
Daily use · storage to release in one tool

Decision

Rebuild as one map everyone works on. Build it myself with AI-assisted development, then hand it to engineering.

Simple Map View: 3D building in production context with review controls
Simple Map View on main: the model in the same context production uses, not a parallel preview.

History

I helped build the original tool in 2022, alongside the engineer who owned the backend: I was the technical artist and 3D lead. Beyond the team, I owned the technical side of 3D graphics on the project: model design and quality, the embedded validation, and the 3D-specific features. From 2022 to 2024 that was a shared build. Late 2024 I stepped back into production leadership, and an engineer took over maintaining it.

By early 2026 the tool needed me again. New surface area had piled up — line details, colorization, landmark icons, and more — on top of QA and release processes that had not kept pace. I came back to rebuild it.

Problem

  • Releases got harder to track as the tool aged. QA and release steps carried years of patched-over process.
  • Features were fragmented across separate, duplicated map implementations. Performance degraded as more got bolted on.
  • Nothing existed for newer map features. Teams either built a one-off tool per need, or tracked 3D models by hand in spreadsheets.

Before and after

Early browser preview of a 3D building from the original 2022 tool
Before. The original 2022 build: proof the pipeline worked, no shared review surface.
Simple Map View in production: a 3D building on the map with the review toolbar and source selector
After. Simple Map View in production: one map, review toolbar, and source switching.

How I built it

I designed the UX and specced the rebuild, then built it with AI-assisted development and prototyping. I went through PR review myself, and deployment went through the team's existing pipeline.

  • Foundation first. Review, lighting, color, compare, and Jira split view attach to Simple Map View. None of them fork another map.
  • Not a PM tool. Releases plug into Jira. The product owns inspection, validation, and promotion, not the roadmap.
  • Ship on main. A side tool would recreate the old split. The new experience had to be the default, not an option.

One path

Intake, validation, review, QA gates, and promotion across development, staging, pre-production, and production, visible as one path.

UX flow of the 3D model lifecycle: upload, review, QA, staging, and production across roles
The path the UI had to support: upload → review → QA → promote. Roles share the map; they do not get separate products.
Model slot with Jira fields, debug coloring by material, and passed validation checks before processing
Server-side checks before processing. Manual review used to be the only gate.

Review, debug, compare

With one shared map, tools could get specific: lighting presets, part color, Jira split view, queue review, and compare against other sources. QA, art, and engineering work on the same surface.

Landmark icon review editor using snapshots from the same 3D models
Icons from the same models. Another consumer of the shared preview path — not a one-off export.

The tool

One product, light and dark. The same parts build every screen.

  • Catalog. Models, slots, landmark POIs, agglomerations.
  • On the map. Review, footprint editing, tile sources.
  • Release. Upload, upload review, releases, KML import.
  • Operations. Statistics, add-ons, access and user rights.

Navigation is five places: Landmarks, Model slots, Releases, Statistics, Add-ons. Environment is a chip on the chrome — dev, staging, pre-production, demo, production — so nobody reviews the wrong build.

Design system

Specified as a system, not a pile of screens. Figma variables are the source of truth, and each variable has a CSS name. 174 variables in 6 collections, with light and dark modes.

  • Foundations. Color primitives, semantic color, type, spacing, shape, elevation, size.
  • Rules. Components use semantic tokens only. Primitives stay inside that layer. Gap, padding, radius, and stroke come from the scales. Every control defines default, hover, focus, and disabled. Minimum target 24×24. Text contrast 4.5:1.
  • Chrome. Buttons, lists, fields, toolbar, panels.
  • Map overlays. Search, light presets (dawn, day, dusk, night), source and style selectors, review queue, camera, split compare.
Map actions
Eiffel Tower eiffel_tower_v3.gltf 1/5
Production 3dbuildings · prod
A demonstration of the components, not the tool itself. Switch the mode, environment, light preset, and camera, or check map actions.

Handoff

Once the rebuild shipped, I moved into the Designer II role: deciding what ships next, in spec and priority, with engineering running the day-to-day. The same track that once maintained the old tool now owns and extends this one.

Outcome

What was a simple map viewer with basic settings is now a full lifecycle tool for 3D buildings: storage, 3D QA, release to the map, work across map features and sources, Jira sync, and the rest of our internal tooling in one place. Validate, review, and release across environments on one map, used daily by 3D artists, QA, and engineers.