Projects & Datasets
An overview of research initiatives, auditing platforms, and community-sourced benchmarks I develop and maintain, spanning algorithmic fairness, region-aware multilingual evaluation, and multimodal AI.
Projects & Research Groups
C-AI — Contextualized AI Research Group @ FaMAF UNC
C-AI (Inteligencia Artificial Contextualizada) is our research group in the Computer Science Section at FaMAF, Universidad Nacional de Córdoba, directed by Dr. Luciana Benotti. We research and develop AI systems situated within their specific socio-cultural and usage contexts, combining participatory bias detection, statistical confidence calibration, multimodal vision-language dialogue, frugal computing, and open evaluation platforms for high-specificity domains such as education and clinical support.
Ethics in AI Team @ Fundación Vía Libre
Fundación Vía Libre is a non-profit civil society organization based in Argentina dedicated to digital rights, technological autonomy, and algorithmic accountability. As Technical Lead of the Artificial Intelligence Team, I direct research on identifying, measuring, and mitigating societal biases and demographic harms in foundation models across Latin America.
EDIA & EDIA 2.0 (Multimodal)
An open-source auditing platform developed to empower civil society organizations, educators, and domain experts to evaluate gender, racial, and regional biases in text embeddings and generative models without needing deep machine learning engineering expertise. Featured as an interactive tutorial at ACM FAccT 2024 in Rio de Janeiro and presented at RightsCon 2023 in Costa Rica.
Datasets & Benchmarks
LACES: Latin-American Community-Sourced Evaluation of Stereotypes
ACL Findings 2026An adaptive, iterative data collection methodology that addresses the critical geo-cultural gap in NLP evaluation by crowdsourcing and validating regional stereotypes and demographic biases across Latin American varieties of Spanish.
HESEIA: Co-Designed Social Bias Benchmark in Secondary Schools
EMNLP 2025 (Oral)A benchmark of 46,499 sentences co-designed directly within secondary school environments across Latin America, engaging 370 teachers and 5,000+ students across 189 schools to audit social biases across curriculum areas. Presented at EMNLP 2025 and featured at the India AI Impact Summit 2026.
INCLUDE-2.0: Region-Aware Multilingual LLM Evaluation
EMNLP 2026 (Main)A large-scale region-aware factual benchmark covering 91 languages from 86 countries (127 language–country pairs) with over 220K curated samples drawn from native regional sources, disentangling linguistic proficiency from locally embedded cultural knowledge.
MultiCrowspairs: Multilingual Stereotype Benchmarking
LREC-COLING 2024Multilingual and cross-cultural evaluation resource addressing the practical challenges of evaluating social stereotypes across global language communities, demonstrating how single-culture US-centric CrowS-Pairs proxies fail to transfer across diverse socio-cultural contexts.
CVQA: Culturally-Diverse Multilingual Visual Question Answering
NeurIPS 2024A global visual question answering benchmark covering 30+ languages across multiple continents, curated by native speakers to evaluate how vision-language models generalize to culturally grounded images and knowledge worldwide.
CaMMT: Culturally Aware Multimodal Machine Translation
EMNLP Findings 2025An evaluation benchmark measuring how effectively multimodal machine translation models understand and preserve culturally specific nuances, visual grounding, and idiomatic expressions across languages.
La Leaderboard: Spanish Varieties & Regional Languages Benchmark
ACL 2025A comprehensive open evaluation suite and leaderboard assessing LLM capabilities across dialectal varieties of Spanish and co-official regional languages of Spain and Latin America.
