International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 13 Issue: 02 | Feb 2026
p-ISSN: 2395-0072
www.irjet.net
MEDEXA – AI-Powered Virtual Healthcare Assistant for Personalized Health Monitoring and Decision Support Ziauddin Mahmood1, Syed Mohiuddin Jeelani Jaffri2, Saniya Abdul Wasay3, Zoya Sohail4,Mohammed Asif5 ¹ Student, Dept. of CSE-DS, Lords Institute of Engineering and Technology, Telangana, India, ² Student, Dept. of CSE-DS, Lords Institute of Engineering and Technology, Telangana, India, ³ Student, Dept. of CSE-DS, Lords Institute of Engineering and Technology, Telangana, India ⁴ Student, Dept. of CSE-DS, Lords Institute of Engineering and Technology, Telangana, India ⁵ Assistant Professor, Dept. of CSE-DS, Lords Institute of Engineering and Technology, Telangana, India ---------------------------------------------------------------------***---------------------------------------------------------------------Key Words: Medical OCR, Healthcare AI, Abstract - Unstructured medical documents, such as Prescription Analysis, Laboratory Report handwritten prescriptions and laboratory reports, continue to create barriers to safe and informed Interpretation, Multi-Engine OCR, Natural healthcare delivery. Illegible handwriting, inconsistent Language Processing, LLaMA, Google Cloud Vision, formatting, and abbreviated clinical terminology Patient Safety. frequently result in patient misunderstandings and medication-related risks. This study presents MEDEXA, a cloud-based intelligent healthcare platform that transforms raw medical images into structured, patientfriendly interpretations using a layered Artificial Intelligence framework.
1. INTRODUCTION Clinical documentation remains one of the most persistent friction points in healthcare delivery. Despite the rapid digitization of hospital systems, a significant proportion of outpatient prescriptions and diagnostic reports are still issued in handwritten or semi-structured formats. These documents often contain abbreviated terminology, nonstandardized notation, overlapping handwriting strokes, and layout inconsistencies that complicate their accurate interpretation. For patients without medical training, understanding dosage schedules, laboratory parameter deviations, or follow-up instructions can become unnecessarily difficult.
The proposed system employs a multistage image enhancement pipeline combined with a three-engine Optical Character Recognition (OCR) strategy to maximize text extraction reliability. A quantitative scoring algorithm evaluates candidate outputs across multiple quality dimensions to select the most accurate representation of document content. The extracted text was subsequently processed using a large language model (LLaMA 3.1-8B) configured with domain-aware prompts to expand abbreviations, normalize corrupted tokens, interpret laboratory values against reference standards, and generate structured JSON outputs for clinical readability.
Medication-related errors and delayed clinical responses frequently originate not from incorrect diagnoses but from miscommunication and document misinterpretation. Handwritten prescriptions may include frequency abbreviations (e.g., BD, TDS, OD) and shorthand drug names that require contextual knowledge. While laboratory reports are numerically precise, they require comparison with reference ranges and clinical reasoning to determine the clinical significance of a value. Thus, the gap between raw medical documentation and patient comprehension remains largely unaddressed at the individual user level.
The platform was implemented using a modern full-stack architecture comprising a Next.js frontend, Node.js backend services, and MongoDB cloud storage, with secure deployment across a distributed cloud infrastructure. Evaluation on a dataset of 200 real-world medical documents demonstrates that the ensemble OCR strategy achieves up to 99.1% accuracy for printed text and 94.8% for handwritten prescriptions while maintaining practical processing latency.
Existing technological solutions tend to operate at two extremes: Institutional electronic health record systems provide structured data management but are costly, infrastructure-dependent, and not universally accessible in outpatient or resource-constrained settings. Conversely, general-purpose conversational AI systems lack the structured extraction mechanisms required for reliable parsing of medical documents. Most research efforts have focused on improving isolated components, such as
By integrating multi-engine OCR, intelligent output selection, and AI-driven semantic interpretation within a production-deployed system, MEDEXA provides an endto-end solution for automated medical document understanding and enhanced patient accessibility.
© 2026, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 706