KIVA

Massachusetts Institute of Technology

KIVA

A scalable AI avatar that delivers personalized, evidence-based vocabulary instruction to close early word gaps in K–4 learners.

Cambridge, Massachusetts, Stati Uniti

USA

Nonprofit

Only 30% of U.S. students reach literacy proficiency, and 43 million adults struggle with basic text understanding. with children from low-income families disproportionately affected, minority backgrounds, and those with learning disabilities. Globally, the crisis is even more stark: an estimated 617 million children are not achieving minimum proficiency. These literacy gaps hinder not only academic achievement but also the ability to make informed decisions and fully participate in civic and economic life. Vocabulary is the strongest predictor of literacy outcomes. Yet early word gaps, particularly among children from disadvantaged backgrounds and those with learning disabilities, emerge before kindergarten and grow wider each year. Interventions are most effective in the early grades, and by fourth grade, when students shift from learning to read to reading to learn, timely support becomes critical. While teacher read-alouds and explicit vocabulary instruction can help, they often benefit students who already have stronger foundational skills. More effective approaches, such as personalized one-on-one or small-group tutoring, are rarely scalable. As a result, many children do not receive the individualized support they need. There is a pressing global need for scalable, high-quality, adaptive instruction that meets learners where they are, especially during this pivotal stage of development.

Our solution is a web-based, AI-based avatar that delivers personalized, adaptive vocabulary tutoring through interactive read-alouds. The avatar can either read the text itself or accompany a child who is reading or listening to an audiobook. Using vocabulary directly from the story, it teaches word meanings through a research-backed approach: providing child-friendly definitions or synonyms, contextual support within the text, and multiple opportunities to engage with the words beyond the story. It then assesses comprehension and collects learning data to support classroom instruction. Under the hood, the platform combines: • Fine-tuned LLMs trained on 200+ hours of elementary-grade tutoring transcripts for age-appropriate dialogue • Child-optimized ASR models for accurate speech recognition and real-time fluency monitoring • Affective computing that senses engagement and adjusts pacing and instruction in real time Instructional strategies follow evidence-based practices—repeated exposure, contextualized usage, and active recall—and adapt continuously to each learner’s performance and emotional state. All interactions occur in a secure, privacy-first environment with voice anonymization and end-to-end encryption. Cost-effective and scalable, the platform is ready to integrate with content providers such as Learning Ally, enabling schools and community programs worldwide to offer individualized vocabulary support at scale.

Our platform targets Kindergarten to 4th‑grade students who are least served by traditional instruction: children from disadvantaged backgrounds, English language learners, and learners with disabilities. Oral language skills at kindergarten entry are the strongest predictors of later achievement because they prepare children to engage with academic language. Yet these students often arrive with limited early language exposure, few opportunities for rich, decontextualized conversation at home, or delayed language development. Well‑documented word gaps emerge before school and widen each year, leaving children increasingly unable to access grade‑level texts or benefit fully from whole‑group instruction. Vocabulary grows through explicit teaching and repeated encounters with sophisticated language. In practice, teacher read‑alouds and whole‑class lessons tend to benefit peers who already possess stronger skills, while one‑on‑one tutoring—most effective for struggling readers—remains too costly to scale. Our AI‑powered avatar delivers adaptive, evidence‑based vocabulary instruction during interactive read‑alouds that adjust to each learner’s needs. We will pilot the platform with U.S. students from underserved backgrounds and with learning disabilities, then expand to reach English and non‑English speakers worldwide, delivering equitable vocabulary growth at scale.