LLM-Driven AutoML for Cross-Lingual Handwritten OCR: Closed-Loop Neural Architecture Search with GPT-5, GPT-4o, and Claude Sonnet 4

Abstract

Handwritten text recognition across diverse scripts presents an enduring challenge in machine learning, as each language and writing system introduces unique visual complexities and demands. Traditional approaches have depended on expert-guided model design and extensive preprocessing, which make it difficult to scale and adapt to new scripts efficiently. In this work, we introduce a pipeline that is fully automatic and cross lingual, using large language models, GPT 5, GPT 4o and Claude Sonnate 4, to independently generate, evaluate, and refine neural network architectures for handwritten optical character recognition. This process requires no manual intervention, domain specific preprocessing, or human selection of models, resulting in a complete end to end automated system.We apply this approach to Arabic, English, and Persian scripts, each representing distinct character shapes and writing traditions, and conduct thirty independent trials for every language. The pipeline consistently discovers efficient models with high test accuracy, achieving average scores above ninety three percent, while also maintaining inference speeds that meet the needs of real time applications. Notably, the system is able to automatically explore a wide range of neural architectures and adaptively select designs that fit the unique requirements of each script, without any explicit guidance from human experts. These results show that large language models can move beyond language processing and act as independent designers for machine learning systems. This enables a scalable, script agnostic, and fully automatic solution for multilingual handwritten text recognition, opening the door to rapid and adaptable deployment of OCR technology across many languages and domains.