I am Don Poole, artist and designer currently investigating how large generative AI models encode and retrieve domain knowledge and what it takes to unlock that knowledge with precision, predictability and consistency. My research builds on 25 years of building simulated real-time environments at scale, with significant contributions as a world designer and builder to award-winning products from Naughty Dog, Sucker Punch, Wevr, and Another Axiom that millions of people experience worldwide.
Latent Space Archeology (LSA) methodology using deep model domain knowledge extraction and synthesis. Text only prompting within an uninterrupted session starting at the street level exterior then "walking" into and up the stairs. Gemini 3.7 Flash with LSA system instruction generated prompt for Gemini 3.1 Flash Image (Nano Banana 2)
I'm currently conducting independent research into spatial constraint systems, testing whether deep access to explicit, cultural, historical and architectural knowledge injected at the prompt level can produce more spatially coherent and visually accurate AI-generated 2D & 3D environments. The work involves deep vibe-coding sessions with Claude Code and Google AI Studio developing an AI-assisted interpretation layer that translates plain-language descriptions into multi-domain grounded prompts that unlock latent knowledge that every-day language prompting or prompt engineering never will.
Modular LSA and Cesium dual pipeline. API dispatch tool using a python LSA domain engine for prompt generation and synthesis with ground truth geospatial and solar data for reconstruction and lighting. Above shows 3 dawn, noon and sunset time of day runs to re-construct lighting to match location and date (09/04/2026).
My foundation is unusual for this space. I studied photography and mixed media at the University of New Mexico under renowned photographers Patrick Nagatani, Tom Barrow, and art historian Eugenia Parry Janis, thinkers who understood that artists don't follow the evolution of their medium, they drive it. That conviction has shaped how I approach every technological shift in my career, from real-time 3D to VR and AR to spatial AI. I don't see generative AI as a disruption to resist, I see it as the next phase of a conversation about how we make, see, and experience the world, one that, for me, started with photography and that continues to evolve as I do.
Early stage Marble world created using LSA domain prompt with Nano Banana 2 image for multimodal input. LSA/Cesium dual pipeline module for Marble 1.1 and 1.1 Plus is underway.