AI-driven educational technologies are expanding rapidly, yet their design rarely reflects the linguistic and infrastructural realities of Deaf learners in low-resource contexts. This qualitative study investigates how teachers, parents, and Deaf students in Bangladesh navigate fragile visual access, inconsistent Bangla Sign Language, and unreliable technology in everyday learning. Through interviews and focus groups with 13 teachers, eight parents, and five Deaf students, we show how small disruptions in sightlines, pacing, or sign clarity can quickly collapse comprehension, making accessibility a condition that must be constantly protected. We identify opportunities for AI to act as access support by stabilizing visual information, ensuring teacher-validated Bangla Sign Language, enabling offline use, and protecting emotional safety when seeking help. We contribute a model of learning continuity that explains how visual, linguistic, and affective stability interact in Deaf education and offer concrete design directions for AI-driven learning tools in the Global South.
ACM CHI Conference on Human Factors in Computing Systems