Keli: A Self-Paced Adaptive Platform for Oral-Aural Learning as a Sociotechnical Equalizer for Hearing-Impaired Students in India

India has one of the world's largest populations of people with significant hearing impairment, yet many schools serving deaf and hard-of-hearing students use oral--aural instruction while access to sustained individualized speech and articulation practice remains uneven. We present Keli, a browser-based AI-enabled learning platform for adaptive oral--aural practice in resource-constrained classrooms. Keli separates measurement, adaptation, and explanation: a deterministic in-browser acoustic module extracts spectral, voicing, energy-band, and duration features to provide interpretable Ling Six Sound feedback with confidence-based abstention; a persistent learner-state model tracks item-level correctness, failures, recency, confusion relationships, and spaced-review history to personalize activity selection and regulate difficulty; and an optional language model converts fixed diagnostic codes into child-appropriate coaching without influencing scores or curriculum decisions. The platform combines lip-reading, animated visualization of otherwise hidden articulatory mechanics, functional communication activities, and game-based retrieval practice across Kannada and English. In a formative three-session feasibility pilot with 15 hearing-impaired students, mean in-app production scores rose from 62.2 to 70.5, 14 students improved on the platform-derived measure, and the number requiring constant adult support fell from 9 to 3. With syllable targets held identical and presentation order counterbalanced, retrieval accuracy was higher in Kannada than in English at every session, a within-subject contrast that isolates familiarity with the language of instruction from phonological difficulty. These preliminary descriptive findings motivate controlled longitudinal evaluation. Keli demonstrates how transparent, bounded AI can support individualized oral--aural practice where specialist time, caregiver capacity, and classroom pacing are limited.