Inteligencia Artificial (IA)
From Aiete to San Francisco: the Donostiarra who trains Artificial Intelligence in gestures
Gianro Compagno
2026-04-29
5 min read
In the heart of global technological innovation, Silicon Valley and San Francisco set a frenetic work pace. There, the '996' culture—working from 9 AM to 9 PM, six days a week—has spread among those seeking to lead the artificial intelligence revolution. In this demanding environment, 27-year-old Xabier Irizar Amuchastegui from San Sebastián stands out.
From San Francisco, Irizar pursues a goal that seems straight out of science fiction: to make digital avatars so realistic that the human eye cannot distinguish them from a real person, thus surpassing the "visual Turing test." "We want avatars to be indistinguishable from humans," Irizar explains from the U.S. West Coast.
His journey began in the Aiete neighborhood of San Sebastián and continued at Santo Tomás Lizeoa. At 18, he moved to Germany to study Engineering Sciences at the Technical University of Munich (TUM), facing the challenge of learning German from scratch and mastering complex subjects like electronics, mechanics, and physics. "The first year was especially tough because I had to translate all my notes to learn the vocabulary," he recalls.
Irizar did not limit himself to studies. He adopted the German tradition of combining academic training with practical experience, participating in exchange programs at the University of California, San Diego, and helping to found TUM.ai, Germany's largest student AI initiative. He also interned at logistics automation companies, which opened doors to Harvard University, where he researched human movement applied to exoskeletons.
Inspired by movies like Iron Man, Irizar became interested in robotics and exoskeletons, dreaming of technologies that allow people to move easily and lift heavy weights. Thanks to the Exponential Fellowship, an organization created by Ignacio Moreno to support young Spanish talents in the U.S., he was able to manage his visa and make the definitive leap into the U.S. industry.
Currently, Irizar is a key figure at Canopy Labs, a fast-growing startup where he was the first employee and which now has about fifteen professionals. The company, founded by young entrepreneurs and backed by investors like Andreessen Horowitz, has secured significant funding rounds.
The technical challenge is enormous: while language and voice models have reached a high degree of naturalness, generating digital faces remains a challenge. "Humans detect any anomaly in facial micro-movements," Irizar points out, aware that perfection in this field is still to be achieved.
Canopy Labs has already launched an open-source voice model and is now focusing its efforts on synthesizing hyper-realistic avatars. Its technology is primarily aimed at businesses, allowing the transformation of texts into videos where digital avatars vocalize messages accurately.
However, creating human replicas raises significant ethical dilemmas. Irizar and his team have set clear limits: they refuse to develop avatars for roles like virtual "therapists," aware of the risks for vulnerable users. Additionally, they strictly control who they sell their models to and apply filters to the texts that avatars can reproduce.
All of this occurs in a context of technological "gold rush," where engineers can earn salaries exceeding half a million dollars, though with no guarantees of job stability. "In two years, the startup either succeeds or disappears," summarizes Irizar, acknowledging the volatility of the sector.
Despite the distance, Irizar closely follows developments in Gipuzkoa, celebrating milestones like the arrival of IBM's quantum computer in San Sebastián and the success of companies like Multiverse Computing, which he considers an example of Euskadi's ability to lead global innovation.