Daniel Koehler

I am an architect who works on one question: how new technology becomes buildable. In the last three years that question has produced a Routledge monograph on generative AI and typology, design projects on Synthetic Building Types, and an atlas commissioned by the Deutsches Architekturmuseum for its own, the ACADIA 2025 Vanguard Paper Award for AI-guided materials discovery, and, most importantly, building materials that did not exist before, made physically with students from compounds an AI had predicted.

What connects these outcomes is a method I call reciprocal grounding: making computational models answerable to the physical world, and the physical world legible to computation, at every scale from crystal structure to city. The research runs in two directions. One measures what models know of the built world: datasets of a million generated images, an atlas of 40,000 places, and the bias analyses that show where a model regresses to a plausible mean. The other designs what models make of it: new building types induced from new materials, developed in atlases that a multi-agent pipeline researches, writes, and adversarially fact-checks.

I teach the way I research. For six years I carried a required lecture course for every fourth-year student at UT Austin and rebuilt it around the technologies my research uses; in Stuttgart I taught fabrication in its most elemental form, students synthesizing predicted compounds with kilns, molds, and the sun. Before academia I spent a decade in practice, including a 30,000 m² timber complex in Nanjing as project architect. Since 2023 I have also served as Chief Scientist of Sengine.ai and, since 2025, as co-founder of LotLens.