Steven Wernke, professor of anthropology, has received a three-year grant from the Schmidt Sciences Humanities and AI Virtual Institute (HAVI) to develop an AI framework that will enable archaeologists to identify specific archaeological features from among millions of ancient buildings across the Andes.
Wernke will serve as principal investigator of the highly collaborative project with co-principal investigators Yuankai Huo, associate professor of computer science; James Zimmer-Dauphinee, postdoctoral scholar; and Parker VanValkenburgh, associate professor of anthropology at Brown University. The team brings together researchers from Vanderbilt and across the U.S., Peru, Chile, Argentina, and Australia.
The $734,574 grant will support the project “Scaling Archaeological Vision: Imperial Landscapes of the Andes through AI-Assisted Survey.” Grants of this size are unusual outside of STEM fields, making the HAVI grant exceptionally competitive: of 501 proposals from 42 countries submitted in 2026, less than 4 percent were selected.
HAVI aims to provide humanities scholars with access to massive AI computing power to accelerate their research and operate at a much greater scale. The premise of the grant is reciprocal: AI tools can help humanities researchers gather large swaths of data in less time and with fewer resources, while humanists can improve AI tools by teaching them the knowledge and cultural context needed to gather the kind of insightful data that pushes their research forward.
Archaeologists have long struggled to fully map and understand the diverse empires that once encompassed vast geographic areas. In the Andes, the rugged, mountainous terrain has previously limited archaeological surveys to isolated areas of a few hundred square kilometers. But the Inka Empire—the largest political formation of the pre-colonial Americas—encompassed about two million square kilometers. The challenge is twofold: to expand the reach of AI-assisted surveys and to see what the surveys detect in greater detail.
“Our earlier work has solved the basic AI detection problem: Our models can now find archaeological features in satellite imagery at continental scale with high accuracy,” Wernke said. “But charting the distributions of features that are diagnostic of specific empires at scale is a needle-in-a-haystack problem that requires richer forms of training and interaction between experts and our models.” The HAVI grant will expand the scope of this research in a way that has not been possible until now.
The project pairs deep expertise in Andean archaeology with cutting-edge research in computer vision and AI. It builds on GeoPACHA (Geospatial Platform for Andean Culture, History, and Archaeology), a web application Wernke developed with VanValkenburgh and Zimmer-Dauphinee, supported by grants from the National Endowment for the Humanities and the American Council of Learned Societies. GeoPACHA enables archaeologists and their student teams to manually survey satellite images and record observations on archaeological site locations. The project then expanded through a National Science Foundation Human Networks and Data Science grant, with Huo joining to lead AI development.
That collaboration produced DeepAndes, the first AI foundation model trained specifically for archaeological remote sensing. Junlin Guo, Vanderbilt Ph.D. student in electrical and computer engineering, led the development of DeepAndes in Huo’s lab. The team then fine-tuned the model with the human-labeled sites from the GeoPACHA platform. DeepAndes is now being deployed across the full extent of the Inka Empire.
While the current DeepAndes AI model autonomously surveys for archaeological structures, it lacks specificity. Several empires expanded across the central Andes over the centuries, each with its own distinct architectural style. The model cannot yet distinguish between them, essentially putting buildings that existed hundreds of years apart on the same map.
The HAVI grant will bring the project into its crucial next phase: developing new AI models able to recognize specific types of key archaeological features. This will enable a global understanding of imperial influence across the diverse social and ecological landscape of western South America. Wernke, Huo, Zimmer-Dauphinee, and VanValkenburgh, along with a team of 15 collaborators, will build and train this new AI framework, which will be called DeepAndes Workbench.
“We will be able to build models to detect particular features such as Wari ceremonial buildings from the eighth century A.D., Inka imperial buildings from the 15th century, and Spanish colonial chapels,” Wernke said. “There may be only a few hundred of these key structures in existence, and even fewer that have been studied archaeologically. This project will enable archaeologists to see something closer to their true distributions across the full extent of these empires for the first time.”
In addition to funding, the HAVI grant includes in-kind supercomputing resources from Schmidt Sciences to realize the full potential of the AI tool. DeepAndes Workbench will function “like an image search engine for the archaeological past,” Wernke said, enabling archaeologists to search for diagnostic imperial features and train future iterations to refine matches even further.
“The result will be the first continental-scale maps of how empires standardized, adapted, and layered their built environments, revealing not only their distinctive ground plans but also how the infrastructure of earlier empires shaped those that followed,” Wernke said. “Since our methods enable the detection of rare features, we think this approach will be successful beyond the Andes, as well.”