SplatGPT Part 2B: Training and Performance
Discover how SplatGPT was trained on curated data, overcame noise, and even predicted meta shifts to optimize competitive Splatoon 3 gear builds.
senior ai systems & infrastructure engineer
Cesar Garza
I build and ship production AI/ML systems end to end — novel algorithms, LLM and agent infrastructure, and the platform around them. I design for the operator: the safe path is the easy path.
recent writing
all writing →SplatGPT Part 2A: Engineering GPT for Set Completion
Introducing SplatGPT, a deep learning model using set-transformers to effortlessly navigate Splatoon 3’s intricate 160B gear-build possibilities
The Deceptive Difficulty of Splatoon 3 Gear Optimization
Splatoon 3 gear optimization hides surprising complexity: 160B possible builds, non-linear stacking, and strategic noise challenging ML models.
SplatGPT: Set-Based Deep Learning for Splatoon Gear Completion
SplatGPT: a set-based deep learning model that completes Splatoon 3 gear builds across a ~160-billion-combination space — architecture, training, and results.