Windows • Character Customization • Production Tooling • Confidential
A Windows-based production tool built to let artists and designers customize 3D characters — facial morphing, hair, skin, and color variations — without touching code. What started as a customization tool became a case study in pipeline engineering after its export step turned out to be the team's biggest production bottleneck.
This project sits alongside my mobile/multiplayer work as proof that my optimization mindset extends to internal tooling, not just runtime gameplay performance.
Facial morphing and body customization are driven entirely through blendshapes, letting non-technical team members produce character variations through simple sliders and presets rather than manual mesh editing.
The original export step processed characters one at a time, synchronously, which meant the production team could only generate around 20 exports in a working session — a hard ceiling on how fast new character variants could reach the game.
I redesigned the export step around coroutine-based processing combined with memory optimization, so exports could be batched and processed without blocking the editor or accumulating memory across the run. This took throughput from 20 exports to 1000+ exports within 5 minutes — roughly a 1900% improvement, and directly unblocked the production team's workflow.
Challenge: Synchronous exports blocking the editor and capping throughput
Each export ran fully synchronously on the main thread, so batch exports meant long, uninterruptible waits and a hard practical limit on daily output.
Solution: Rebuilt the export step as a coroutine pipeline, processing characters incrementally instead of blocking, and adding memory optimization to prevent accumulation across large batch runs.
Challenge: Keeping the tool usable for non-programmers
The performance fix couldn't come at the cost of usability — artists still needed a simple, slider-driven interface for character customization.
Solution: Kept the customization UI untouched and isolated the optimization entirely inside the export pipeline, so the artist-facing workflow stayed exactly the same while throughput improved underneath it.
🔒 This project is confidential and its source is not publicly available. Details above are shared with permission for portfolio purposes.