The Future of Education

Dyslexia is a complex learning difficulty that impacts reading, spelling, and can manifest in the visual–motor aspects of handwriting. Early detection is critical but can be hindered by subjective or overly narrow assessments. This paper presents a dual-model pipeline that integrates (1) Gemini 2.0 Flash Exp for image-based handwriting analysis and (2) OpenAI O1 for text-based spelling assessment, with a transparent logistic regression classifier. On a balanced dataset of 100 handwriting samples (50 dyslexic, 50 non-dyslexic), the pipeline achieved near-perfect classification on a 20-sample validation subset. We show how combining morphological and text-driven features into a single representation allows logistic regression to produce a continuous dyslexia probability, enabling threshold-based categorization. While future studies with larger samples and additional modalities (e.g., reading fluency, eye tracking) are needed, this work sets a foundation for multi-cue, interpretable dyslexia screening methods.