PragyaDoc reads medical documents in 22 Indian languages for rural healthcare

PragyaDoc, a new universal document intelligence framework from researcher Jagpal Singh Jhala, addresses a critical accessibility gap in India's healthcare system: while most medical documentation is written in English, the majority of patients—especially in rural areas—cannot read it. The paper, posted on arXiv, describes a four-layer pipeline designed to extract, structure, and reason over multilingual medical documents across all 22 of India's official languages. This is a major step toward democratizing medical information for non-English-speaking populations.

The framework's architecture is notable for its layered approach. First, a parallel ensemble OCR layer captures text from varied document formats and scripts. Next, a geometric-lexical fusion layer combines spatial layout analysis with linguistic features to preserve the meaning of complex medical forms. A deterministic domain structuring layer then organizes the extracted data into standardized medical categories, ensuring reliable output. Finally, a dual-LLM medical reasoning and localization layer uses two language models to interpret clinical content and translate it into local languages and contexts. This design aims to work reliably even in low-resource settings where computational power is limited. For nurses, community health workers, and patient families, PragyaDoc could mean the difference between being excluded from critical diagnoses and actively participating in healthcare decisions.