As educational technology platforms become more and more commonplace in education, it is critical that these systems work well across a diverse range of student sub-groups. In this study, we estimated the effectiveness of MAP Accelerator; a large-scale, personalized, web-based, mathematics mastery learning platform. Our analysis placed a particular focus on students from historically under-resourced groups. Our sample comprised 181K students in grades 3-8 from 99 school districts across the United States, 52% of whom attended schools where the majority of the students are eligible for free or reduced price lunch (a proxy for high-poverty schools). Using a combination of system logs/platform telemetry data, standardized assessments, and publicly available school/district data, we estimated the causal effect of a year-long supplemental math intervention on students' standardized mathematics growth outcomes via a quasi-experimental design with controls and a modification of the difference-in-differences framework. Students who used the platform as recommended (30+ min/wk) during the (COVID-19 disrupted) 2020-2021 school year had math growth scores 0.26 standard deviations higher on average than similar students who used the platform for <15 min/week. Further, positive benefits of the platform were observed across all genders, race/ethnic groups, and school poverty levels, but were not as large for districts with greater than 20% English language learners. Further analysis revealed that these students were predominantly Hispanic, and tended to make less progress on fewer skills than their peers given the same amount of practice time. We discuss the implications of these findings, and potential avenues to ensure more equitable outcomes for these students in the future. [For the full proceedings, see ED623995.]