ABSTRACT Numeracy difficulties among young learners are still one of the main persistent issues faced by Nigeria’s basic education system, with high number of pupils in public primary schools unable to perform basic arithmetic exercises. Already vulnerable young learners, particularly those in poor-resourced, rural, and communities faced with conflicts, are severely affected due to large class sizes, teacher shortages of teachers and the lack of individualized instructional support. Conventional, one-size-fits-all numeracy instruction has proven insufficient in tackling the heterogeneous learning demands of these children, as it is largely uniform in pace and content and incapable of responding to individual misconceptions in real time. This study proposes and evaluates an AI-powered adaptive learning platform that integrates learner profiling, diagnostic assessment, and machine learning-driven content sequencing to deliver individualized numeracy intervention for at-risk early learners. Adopting experimental research design, the study collected and analysed structured learner interaction data from a stratified sample of at-risk early learners drawn from rural and urban school contexts, comparing outcomes with a non-adaptive control group. The platform adopts classroom developed diagnostic findings, response-time logs, and error-pattern analysis to continuously calibrate task difficulty and provide arrangements, real-time feedback to learners. Adaptive algorithms, including rule-augmented machine learning models and mastery-based sequencing techniques, were put in place and assessed using indicators such as numeracy mastery gain, rate of engagement, error-reduction rate, and time-to-mastery. The study is guided by five research questions and corresponding hypotheses. The data shows that AI-powered adaptive platforms notably do better than static, non-adaptive numeracy instruction in accelerating mastery among vulnerable pupils while maintaining greater engagement and less frustration-related disengagement, and that the study generalise across rural, semi-urban, and urban learner groups. All null hypotheses of no difference were rejected. The developed foundation roots adaptive learning in a continuous intervention cycle which includes diagnosis, personalisation, feedback, and progress monitoring. This research concludes that AI-powered adaptive learning offers a quantifiable and context-sensitive remedy for reinforcing foundational numeracy end result, lowering learning poverty, and supporting equitable basic education delivery in Nigeria, and recommends its integration into current national learning-recovery programmes. Keywords: Artificial Intelligence, Adaptive Learning, Numeracy Intervention, At-Risk Learners, Basic Education.