AIRE: AI Reading Enhancer for Personalized Decodable Texts in Early Literacy

• In Event: Innovating K–12 Teaching and Learning with Generative AI: Insights From IES-funded National R&D Centers

Abstract

Objectives or Purposes
This study introduces AIRE (AI Reading Enhancer), an AI-powered literacy tool that uses teacher inputs to generate personalized decodable books with integrated real-time assessment and reading support for K–2 students. AIRE addresses the critical gap in high-quality decodable books by combining large language models (LLMs) with teacher-defined instructional goals and student-specific profiles to enhance students’ decoding, fluency, and comprehension capabilities.

Perspective(s) or Theoretical Framework
AIRE is grounded in the science of reading, culturally sustaining pedagogy, and human-AI co-
learning frameworks. Drawing on cognitive load theory, it scaffolds instruction through
decodable text tailored to students’ dialects, experiences, and learning profiles (Sweller, Ayres & Kalyuga, 2011). The system uses a teacher-in-the-loop design to preserve instructional control while incorporating adaptive AI features. Its real-time assessment loop reflects sociocognitive theories of feedback and formative learning.

Methods
AIRE employs a four-step content creation and assessment workflow: (1) story text generation guided by phonics goals and learner profiles, (2) teacher review for educational alignment, (3) multimodal content generation (illustration, narration, vocabulary), and (4) final teacher approval. During student reading, AIRE uses automatic speech recognition (ASR) and phonics alignment models to evaluate oral fluency and decoding accuracy. It dynamically generates assessment questions via LLMs to probe comprehension, vocabulary, and decoding understanding. These responses are used to trigger just-in-time feedback through the AIRE Text Enhancer module.

Data Sources
The prototype development is based on (1) a curated benchmark of phonics-aligned vocabulary from DIBELS-8 and UFLI, (2) an expert-designed rubric for evaluating text quality, and (3) focus group feedback from K–2 educators. Planned data collection includes ASR-based audio logs and student response data to LLM-generated assessment items.

Results
Preliminary evaluations demonstrate that AIRE generated texts outperform baseline children’s
books in aligning with phonics and vocabulary benchmarks. Teachers rated the content as
linguistically appropriate, culturally responsive, and instructionally effective. Early tests of the
ASR phonics analysis pipeline can identify student decoding errors and comprehension gaps,
which enables adaptive feedback and teacher insight. Teachers highlighted the benefit of
embedded assessments and valued the real-time analysis of student reading as a tool for
differentiation.

Scientific/Scholarly Significance
This research contributes to the field of AI in education by offering a scalable, ethically
designed, and instructionally grounded system for early literacy instruction. AIRE fills a critical
gap in decodable book reading and learning by integrating real-time assessment, LLM-
generated instructional content, and teacher-guided feedback. It provides a model for human-
centered AI that preserves teacher agency, ensures cultural and pedagogical alignment, and
supports equitable, personalized learning in K–2 classrooms.

Authors

• Zhaohui Li, Pennsylvania State University

• Qingxiao Zheng, University at Buffalo - SUNY

• Dancheng Liu, Buffalo State - SUNY

• Tanya M. Christ, East Carolina University

• John Z. Strong, University at Buffalo - SUNY

• Laura S. Tortorelli, Michigan State University

• X. Christine Wang, University at Buffalo - SUNY

• Abeer Alwan, University of California - Los Angeles

• Dilek Hakkani Tur, University of Illinois at Urbana-Champaign

• Jinjun Xiong, University at Buffalo - SUNY