Unreal / DevOps Engineer [Contract]
Nov 2024 – Jan 2026
One of three Unreal engineers on the project. Led the charge on critical features such as pixel streaming, devops, automation, deployment, and embedded chromium. Played critical roles in turn-based behavioral networking systems, UI, audio processing, lipsync, and iterating on agent alignment.
Helped lead the team from the first investor demos through to the company's acquisition by Luma AI.
Overview
I was one of the three Unreal engineers on the project. The team was veteran talent, including ex-Sony VFX and character artists and animators, and I worked directly under Naughty Dog cofounder Andy Gavin. Our small team focused on the visual "puppet" side of the AI. I joined and soon developed/owned our entire build, deployment, and validation pipeline with QA. Diagnosed issues in packaging as they arose. Created animator tooling and overhauled tooling UI to be procedural throughout our expanding feature set.
We then quickly pivoted into pixel streaming to target mobile deployment, hosting our builds on AWS EC2 GPU Linux servers. I led the charge on deployment procedures and automation for this new system, including gathering traces and logs on target hardware to ensure a consistent 30fps on a reasonable GPU.
Patched a bug found in these traces at one point that improved our CPU tick time 50x, letting us hit our CPU targets easily. Consistently tested for reproducible hitches and diagnosed them to ensure a stable framerate.
Soon I expanded into behavior programming, acting as one of the few full stack engineers across backend and frontend, designing our presentation system's behavior and Unreal implementation with Chromium to support more flexibility with our web deployment team. From there I kept moving into the cognitive behavioral, AI prompting side of the system.
It's a startup, so a generalist mindset and adaptability are huge. I was always eager to throw my hat in the ring and try something new.
Experience Highlights
- Created processes for deploying Naughty Dog-quality visuals on Linux AWS instances that were streamed to users at stable 1080p 30fps. 99.9% uptime, with ability to gather traces and logs on deployed hardware.
- Ensured high quality, low-latency video stream from Unreal while working in inconsistent network conditions.
- Owned deployment process to production and ensured stability for consistent weekly investor demos.
- Developed artist pipeline for ensuring and correcting visual parity between Windows and Linux. Built automated build, cook, and deployment pipelines for Windows, Mac, and Linux, reducing manual release overhead.
- Diagnosed LLM state machine failures that users encountered and resolved prompt alignment issues.
- Implemented Unreal to LLM behaviors, presentation mode, and conversation-state fixes for agent interactions.
- Built synthetic data generation in Unreal to produce training sets for the ML team.