The objective of this engineering project was to design and evaluate a software system that improves the semantic accuracy of text extracted from handwritten and printed documents. Conventional Optical Character Recognition ( OCR ) systems frequently fail to accurately interpret complex or highly cursive handwriting, resulting in unreadable or misleading output. To address this problem, I engineered and developed a web - based application called DigiScribe which uses an agentic AI framework. The system preprocesses uploaded images by reducing noise, performs initial text extraction using traditional OCR techniques, and assigns confidence scores to recognized text segments. Next step is a context - aware refinement pipeline which selectively improves low - confidence segments by incorporating user - defined contextual information and character - level constraints which are optional, while preserving the original text order and high - confidence outputs. System performance was evaluated by comparing unrefined OCR outputs with context - refined outputs across multiple test samples. Evaluation criteria included OCR confidence scores and human readability assessments. The engineered system consistently produced more accurate and readable text, particularly in regions containing ambiguous handwriting. Importantly, high - confidence segments remained unchanged, ensuring stability and minimizing unintended modifications. This project demonstrates that an agentic, context - aware refinement architecture can significantly enhance OCR performance. The resulting system has practical applications in document digitization, educational material preservation, and accessibility of handwritten content. The modular design of the system allows for future extension and optimization.