When Teachers Program the AI: Empowering Educators to Create Context-Aware AI Learning Tools for the Roma Community in Spain
Roma students in Spain experience persistent educational disadvantage, reflected in early learning gaps, high rates of grade repetition, and elevated levels of early school leaving. These disparities emerge early and widen over time, limiting access to higher education, formal employment, and full social participation. Fundación Secretariado Gitano (FSG), the largest Roma NGO in Spain, operates after-school support programs intended to mitigate these inequalities, but educators in these settings often work with students who differ widely in academic levels, learning needs, and socio-emotional vulnerability, without practical tools to tailor instruction effectively. As a result, many students remain at high risk of continued educational exclusion despite participating in existing support structures. The central research question is: can empowering teachers with the autonomy and skills to customize AI tools for their students’ needs improve outcomes for vulnerable learners?
The intervention seeks to improve learning, engagement, self-efficacy, and effort among Roma students by giving educators a practical way to personalize academic support within FSG after-school centers. Teachers in treatment centers receive training, coaching, and access to PlayLab, a no-code platform that allows them to design or adapt AI tutoring chatbots without programming. These tools are intended to align with students’ academic levels, needs, and cultural identities, while keeping teachers in control of pedagogy and instructional choices. During after-school sessions, students use the chatbots as part of regular group tutoring alongside the teacher’s in-person guidance. The aim is not to replace teachers, but to extend their reach and make support more adaptive and relevant.
The project will be evaluated through a cluster-randomized controlled trial. The unit of randomization is the teacher-block, defined as one or more educators who co-teach or share students within the same after-school group, in order to reduce spillovers among teachers working closely together. The study will include around 120 teacher blocks across 56 FSG centers in 13 regions of Spain, serving about 1,700 students. Treatment teacher-blocks will receive training, coaching, and access to PlayLab, while control teacher-blocks will continue with business-as-usual after-school support. The main analysis will estimate intent-to-treat effects, with heterogeneity pre-specified by baseline achievement, gender, and teacher digital skill.
The evaluation combines several sources of data to measure both outcomes and mechanisms. Student learning will be assessed at baseline and endline using standardized adaptive assessments in math and reading, together with aligned English assessments. Student surveys will capture engagement, effort, motivation, self-efficacy, and belonging, while teacher surveys will measure self-efficacy, stress, time use, technology use, and personalization practices. Platform log data from PlayLab will document chatbot use, iteration depth, scaffolding features, tone, cultural references, and student interaction patterns. These measures will be complemented by FSG and school administrative data, including attendance, grades, and student background characteristics.
The theory of change operates through two reinforcing channels. The first is instructional: teacher-designed AI chatbots can provide more consistent scaffolding, individualized pacing, adaptive questioning, and targeted feedback, making it easier to support students with different academic levels within the same group. This is particularly relevant in settings where educators may lack the time or subject-specific knowledge to personalize instruction continuously. The second is identity and
engagement: because teachers can embed familiar language, affirming tone, and Roma cultural references into chatbot design, the intervention may strengthen students’ sense of belonging and reduce stigma and disengagement. Improvements in motivation, confidence, and persistence are expected to reinforce the direct instructional gains.
The project has broader policy relevance beyond the specific case of Roma students in Spain. It can generate evidence on whether a low-cost, teacher-led use of AI can strengthen foundational skills, engagement, and inclusion among students facing persistent educational disadvantage. In doing so, it speaks to wider policy debates in Spain and Europe on remediation, educational inequality, and the role of AI in education. More generally, the study can inform the design of scalable interventions for hard-to-reach populations while offering practical guidance on how to integrate AI in ways that complement, rather than displace, teachers.