Level select

Level 02Downtown Data Center

01 / 09

01 Live-service simulation · CHPC

Internal tool

Downtown Data Center

A low-poly, high-readability virtual twin of the University of Utah's downtown data center. Technicians used it for remote monitoring, debugging and training, and it streams to the browser, so there's nothing to download.

Role
Game Designer (Generalist) · official title Graphic Software Developer
Team
Two-person core team · CHPC, University of Utah
Timeline
Dec 2023 – Jul 2026 (full time and on site from May 2025)
Built with
Unreal Engine 5.3 · Maya · Pixel Streaming

Internal use only. The simulation holds sensitive infrastructure data, so there's no public build; this deck shows recordings and screenshots.

Low-poly render of rows of black server racks in a dim data center with a red-lit wall.

02 The question

How do you turn real-time infrastructure data into a usable 3D interface?

  1. ProblemChecking racks, alerts and jobs relied on physical methods that slowed staff down.
  2. DesignA low-poly, alert-first space: loud alerts, lit server fronts and a clear sensor-data UI.
  3. SystemOne data table spawns 200+ servers, SQL pipelines feed inventory and rack data, and Pixel Streaming delivers it in the browser.
  4. UserI surveyed staff, mapped the areas they rely on most, and folded volunteer testers' feedback into releases.
  5. ResultTechnicians used it for remote monitoring, debugging and training. Internal use only.

I owned

  • User research and requirements.
  • 200+ server models in Maya and the data-table spawning system.
  • UI and UX for the sensor data.
  • Engine moves (5.3 to 4.23 and back) and Pixel Streaming on Linux.
  • Pipeline documentation and onboarding.

I worked with

  • One other developer: a two-person core team.
  • Engineering, for the rack-mapping data and inventory portal.
  • Data center staff, as users and volunteer testers.

Internal use only: the simulation holds sensitive infrastructure data, so there is no public build.

03 Watch

Two builds, about eight months apart

A walkthrough from September 2024 and one from May 2025.

September 2024 May 2025

04 User

A tool, not a replica

The goal wasn't a photorealistic copy of the data center. It was a clear, functional tool that staff would actually use instead of falling back on older, physical methods.

  • Surveyed staff on the metrics they track and how they look for information, then mirrored that in the simulation.
  • Mapped the physical data center and prioritized the areas staff rely on most.
  • Watched where physical methods slowed people down: clutter, poor visibility, and hard-to-spot alerts.
  • I observed user sessions and collected usage data to fix bugs in the live build.

Clarity through design

  • Low-poly environment to cut visual noise and keep performance high.
  • Clear UI/UX built around the end user and iterated on feedback.
  • Visually loud alerts that stand out even across the room.
  • Purposeful lighting that highlights server fronts for quick recognition.
  • Browser delivery so access works on any OS with no installs.

05 System · Data pipeline

200+ servers, spawned from data

Every server model has its own front-panel texture, sourced from online research and on-site reference photos, and is scaled to real-world dimensions from the inventory team.

I replaced 200+ one-off server Blueprints with a single data-table-driven spawning system keyed on name, texture, material and size, structured to cut RAM use. SQL pipelines feed inventory and rack data into the simulation, and a plain-language guide lets new team members add servers on their own.

Spreadsheet of server models and their rack-unit sizes.
The legwork. Inventory sheet.
Input Unreal data table listing server names, materials, textures and sizes.
INPUTA data table with each server's name, texture, material and size.
Output A row of racks filled with distinct, textured server models.
OUTPUTDistinct servers loaded into their assigned rack slots.

06 Design · UI / UX

Sensor data you can read at a glance

Clicking a rack opens its details: the chassis and nodes, alert status, running jobs, power use and connected outlets, with a color key for alerts and jobs.

The details HUD laid out in Unreal's widget editor, with sections for chassis, nodes, running jobs, alerts and outlets.
FIG. 01The details HUD, laid out in the widget editor.
Runtime rack details screen for rack S-13: a rack model with temperature icons, chassis info, node buttons and a no-issues alert status.
FIG. 02The same HUD at runtime, with temperature readouts on the rack.

07 System · Deployment

Getting it into a browser, on any OS

Technicians shouldn't have to install anything. After evaluating NVIDIA WebGPU, HTML5 export and Pixel Streaming, the project took three engine and platform moves.

  1. Start

    UE 5.3

    The simulation began in Unreal Engine 5.3.

  2. HTML5 release

    UE 4.23

    I led the migration: moved assets, optimized scenes and built supporting Blueprints to ship an internal HTML5 build.

  3. Final setup

    UE 5.3 + Pixel Streaming

    Back on 5.3, streaming from a dedicated Linux server once that path proved more scalable.

Multi-instance

Architected two isolated Pixel Streaming instances on separate ports on Windows.

Linux

Migrated the build to Rocky Linux 9, cross-compiling with Clang 16.0.6 and fixing the compiler and toolchain errors that came with it. I wrote C++ for the cross-platform deployment.

Prototype

An AR overlay in UE 5.4 that shows rack data when staff point a phone at a physical rack.

08 Creation process

From site visits to a scalable pipeline

  1. Research

    Mapped workflows & users

    • Reused assets from an earlier 3D touring project to move faster.
    • Surveyed staff and mapped the areas they rely on most.

  2. Core systems

    Assets & data

    • Modelled 200+ low-poly servers and pods in Maya to precise measurements, down to screw points.
    • Tested the engineering team's rack-mapping data and caught overlaps and mismatches with the floor.

  3. Expansion

    Usability & delivery

    • Presented to the department, recruited testers and folded their feedback into releases.
    • Designed UI for sensor data across several areas of the simulation.

  4. Scale

    Docs, polish, mentoring

    • Wrote pipeline documentation for onboarding, and I mentored new team members.
    • Fixed UV issues and added models for environmental detail.

09 Level complete

← Back to level select