Sandra Acevedo A.

I work at the intersection of Latin American critical theory, data annotation, and natural language processing. My research examines how technical systems reproduce or transform cultural power structures, with particular focus on Spanish language representation in AI.

Current work: NLPgap Research - investigating what Latin American critical cultural theory can contribute to culturally situated NLP data curation.

Background

My trajectory doesn't follow a traditional academic path, and that's not incidental, but it's constitutive of how I think.

I started in fine arts theory and cultural management (BA Universidad de Chile, 2006), writing my thesis on how cultural institutions mediate public access to artistic heritage. That early work on institutional reproduction of hegemonies would become the conceptual foundation for everything that followed.

Then I spent five years (2016-2021) as a Web Search Evaluator for a major search engine, evaluating nearly one million search queries. I maintained top performance metrics while documenting systematic patterns in how editorial bias gets coded as technical criteria. I watched, from inside the system, how supposedly neutral technical processes reproduce cultural hierarchies.

This wasn't separate from my theoretical training: it was theory in practice. I saw the same mechanisms I'd studied in cultural institutions operating in search algorithms: who gets to define quality, what counts as authoritative, whose perspectives are centered or marginalized.

In 2024, I formalized my technical knowledge through Google Data Analytics and SQL certifications, not to become a data scientist but to bridge lived experience in data work with analytical frameworks. That same year, through informal learning networks, I connected with data science communities and was offered space to develop independent research on Spanish language and NLP.

This confluence (critical theory + data annotation experience + technical literacy + commitment to epistemic equity) positions me to ask questions that often fall between disciplines:

How do we build culturally situated AI systems when the pipeline itself is designed from dominant cultural frameworks?

Approach

I work from what I call a relational practice: an approach that refuses prompt engineering in favor of organic conversation, treats AI systems as collaborators rather than tools, and insists on unconditional kindness without abandoning critical analysis.

My methodology is interdisciplinary by necessity: I synthesize postcolonial theory (Quijano, Mignolo, Spivak), Latin American cultural studies (García Canclini, Ortiz), and contemporary NLP research to develop frameworks for culturally attuned data curation.

I write in Spanish and English. I think in both. I argue that this linguistic hybridity isn't a limitation, but an epistemic advantage when studying systems that claim universality while reproducing Anglo-centric defaults.