AI-Powered Healthcare Assistant with Drug Interaction Detection and Emergency Support

Abstract

Modern healthcare delivery continues to be con-strained by fragmented digital tools, delayed diagnosis, unmoni-tored drug-drug interactions (DDIs), and insufficient emergency response mechanisms, particularly for users without direct spe-cialist access. This paper presents an AI-Powered Healthcare Assistant—an integrated, secure, web-based platform that unifies conversational symptom analysis, hybrid disease-risk classifi-cation, knowledge-graph-based drug interaction detection, and geolocation-enabled emergency support within a single appli-cation. The system combines a GPT-based NLP pipeline for symptom interpretation, a hybrid rule-and-scoring risk engine classifying patient state into Low, Medium, High, and Emergency tiers, and a Neo4j biomedical knowledge graph modelling drugs, diseases, genes, and proteins to flag Mild, Moderate, or Severe medication interactions. Built on React.js, Node.js/Express.js, MongoDB, and Neo4j, and secured through JWT authentication, bcrypt password hashing, and role-based access control (RBAC), the platform achieved a validated disease-prediction accuracy of 87.3% and end-to-end response latency suitable for interactive clinical decision support. Grounded in an evaluation spanning healthcare microservices, explainable AI, clinical NLP, graph neural networks, and healthcare data security, the system demon-strates that a single, well-secured platform can meaningfully bridge the gap between AI-driven diagnostic guidance and actionable, real-time patient care.

Introduction

The text presents an AI-powered integrated healthcare assistant designed to improve healthcare accessibility, patient safety, and emergency response through artificial intelligence, machine learning, natural language processing (NLP), and biomedical knowledge graphs.

The system addresses major healthcare challenges, including delayed diagnosis, overlooked drug-drug interactions, limited access to medical specialists, and fragmented healthcare services, particularly in rural and semi-urban areas.

A key focus is drug-drug interaction (DDI) detection, where graph neural networks and biomedical knowledge graphs are used to identify direct and indirect interactions between medications. Unlike traditional databases, this approach can analyze complex relationships among drugs, diseases, genes, proteins, and biological pathways.

The proposed platform integrates symptom analysis, disease prediction, medication safety assessment, patient medical history, and location-based emergency assistance into a single conversational interface. It uses a hybrid risk assessment engine that combines rule-based emergency detection with machine-learning models to classify health risks as low, medium, high, or emergency.

The system is built using a microservices architecture, with React and Vite for the user interface and Express.js for API management. Neo4j is used to represent biomedical relationships, while security mechanisms such as JWT authentication, bcrypt password hashing, role-based access control (RBAC), and audit trails protect sensitive patient information.

The research also highlights the importance of explainable AI, scalable clinical computing, and healthcare data privacy. According to the text, the system achieved 87.3% disease prediction accuracy and API response times below 200 milliseconds, alongside reported OWASP compliance.

Conclusion

This paper presented an AI-Powered Healthcare Assistant that unifies conversational symptom analysis, hybrid risk classification, Neo4j knowledge-graph-based drug interaction detection, and geolocation-enabled emergency response within a single, secure architecture. By validating the system against real-world clinical test cases, the platform achieved an 87.3% disease prediction accuracy, 99.4% critical symptom recall, and robust sub-200ms non-AI API response times while maintaining full HIPAA-aligned data governance. Future research will focus on: 1) GNN Zero-Shot Interaction Prediction: Integrating an EmerGNN-style graph neural network to predict inter-actions for newly released molecules lacking established graph edges. 2) Differential Privacy and Federated Learning: Imple-menting DP-SGD and federated training routines across decentralized hospital datasets to continuously update risk scoring without aggregating raw patient data. 3) Multilingual Conversational NLP: Fine-tuning multi-lingual language models (e.g., IndicBERT, mBERT) to support regional dialects. 4) IoT Telemedicine and Wearables: Ingesting real-time vital telemetry from wearable sensors (SpO2, heart rate variability, ECG) to enable continuous, dynamic health monitoring.

References

[1] van der Vegt and S. R. Abbas, “AI-Driven Microservices Architecture for Scalable Healthcare Data Platforms,” MDPI Electronics, vol. 13, no. 17, pp. 3401–3418, 2024. [2] A. van der Vegt and S. R. Abbas, “Layered Microservices Intelligence Architecture for Smart Healthcare Systems,” JAMIA, vol. 14, no. 9, pp. 812–826, 2024. [3] P. Melke and K. Zhioua, “GraphRAG and Native Vector Indexing for High-Performance Knowledge Graph Systems,” in Proc. NODES Conf., 2024, pp. 102–115. [4] Q. Abbas, W. Jeong, and S. W. Lee, “Explainable AI in Clinical Decision Support Systems,” Healthcare (MDPI), vol. 13, no. 17, pp. 1920–1938, 2025. [5] S. Liu et al., “Leveraging Explainable Artificial Intelligence to Optimize Clinical Decision Support,” JAMIA, vol. 31, no. 4, pp. 580–594, 2024. [6] M. Alonge and O. Isreal, “Hybrid AI Models for Improving Decision Quality in Health and Enterprise Applications,” ResearchGate Technical Review, 2025. [7] J. Lee et al., “BioBERT: A Pre-trained Biomedical Language Represen-tation Model for Biomedical Text Mining,” Bioinformatics, vol. 36, no. 4, pp. 1234–1240, 2019. [8] E. C. Chianumba et al., “NLP Models for Extracting Healthcare Insights from Unstructured Medical Text,” IJAMRS, vol. 4, no. 6, pp. 312–325, 2024. [9] Y. Zhang et al., “EmerGNN: Emerging Drug Interaction Prediction,” [10] Springer Nature, pp. 1–16, 2023. [11] Heterogeneous Knowledge Graph Representation Team, “HeteroKGRep: Heterogeneous Knowledge Graph Representation,” IEEE / arXiv, 2023. [12] S. Hoffman and A. Podgurski, “Securing the HIPAA Security Rule,” Case Western Reserve University, Working Paper, 2006. [13] M. Abadi et al., “Deep Learning with Differential Privacy,” in Proc. ACM CCS, 2016, pp. 308–318. [14] R. S. Sandhu et al., “Role-Based Access Control Models,” IEEE Computer, vol. 29, no. 2, pp. 38–47, 1996. [15] ISO TC 212, “Medical Laboratories – Requirements for Quality and Competence (ISO 15189:2022),” International Organization for Stan-dardization, Geneva, 2022.

Copyright

Copyright © 2026 Savara Gnan Prakash, Diwakar Nallappagari, Gangana Jashwanth Reddy, Veduruparthi Raja Krishna Deepak, Sulake Shanmuka Rao, Ritu Agarwal. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.