Interpreting Medical Records with Generative AI: A Patient-Facing Chat Application Using Google Gemini

Abstract

In the age of digital health, patients are increasingly gaining access to their personal medical data, including lab test results, vital signs, and diagnostic reports. However, the technical and clinical complexity of this information often renders it difficult for non-medical users to interpret, potentially leading to confusion, misinformation, or anxiety. This research presents the development of an AI-powered web application that assists users in understanding their health data through natural language interaction. The system utilizes Google’s Gemini generative AI model to process textual medical information and provide context-aware answers to user queries. The application, developed using the Streamlit framework, supports both direct text input and file uploads in PDF or TXT format. Extracted data is sent to the Gemini API using a structured prompt design that frames the model as a medical data assistant. The model’s output is constrained with instructions to prioritize accuracy, avoid conjecture, and issue disclaimers discouraging self-diagnosis. Core components of the system include file parsing using PyPDF2, secure API key management via dotenv, asynchronous API interaction, and a chat-like user interface that maintains dialogue context using Streamlit’s session state. This project demonstrates a practical and ethical implementation of generative AI in consumer-facing health technology. By enhancing the accessibility and interpretability of medical records, the application empowers users to become more informed participants in their healthcare journey, while reinforcing the importance of professional consultation for clinical decisions.

Introduction

With the widespread use of Electronic Health Records (EHRs) and online patient portals, patients now have unprecedented access to their medical data. However, due to the complexity and clinical jargon, most people—88% of U.S. adults—struggle to understand this information, which can lead to anxiety, misinterpretation, and poor health decisions.

2. The Role of Generative AI

Generative AI, especially Large Language Models (LLMs) like Google’s Gemini, offers a way to bridge this gap by interpreting complex health data in plain language.

Key capabilities of LLMs include:

• Natural language interaction for everyday questions.

• Contextual understanding of medical documents.

• Adaptive explanations based on user knowledge.

• Personalized responses tied to specific health concerns.

• Multi-format support (lab reports, clinical notes, etc.).

? Benefits:

• For Patients: Increases health literacy and informed decision-making.

• For Clinicians: Reduces routine explanation burden.

• For Healthcare Systems: May reduce unnecessary visits caused by confusion.

?? Challenges:

• Risk of hallucinations (inaccurate responses).

• May miss patient-specific details.

• Needs clinical oversight and ethical safeguards.

3. Research Objectives

This study explores the feasibility and impact of using generative AI to help non-experts understand their personal medical data. Goals include:

1. Building a user-friendly conversational AI interface.

2. Testing Gemini's ability to handle real medical documents.

3. Developing prompt engineering strategies to guide safe responses.

4. Addressing technical and ethical challenges.

5. Identifying future integration paths with health systems.

Note: The goal is not diagnosis, but enhancing data accessibility for patients.

4. Literature Review Summary

A survey of 10 key studies reveals rapid progress in AI-powered medical chatbots, including:

• Med-Bot, ChatDoctor, and Clinical Camel: Fine-tuned LLMs for domain-specific medical dialogue.

• HuatuoGPT and CareBot: Models trained on real-world doctor-patient interactions.

• eHealth Assistant: Privacy-preserving, secure AI using decentralized protocols.

• Studies highlight improved patient support, but consistently warn of the need for:

  • Ethical safeguards

  • Robust training data

  • Human oversight

5. Methodology Overview

????? System Architecture (Three Layers)

1. Presentation Layer: Streamlit-based web app for interaction and guidance.

2. Processing Layer: Extracts and formats medical data (from PDFs or text).

3. Intelligence Layer: Interfaces with Gemini API for AI-driven interpretation.

???? Data Flow

1. User uploads/pastes medical data.

2. System extracts content.

3. User enters a health-related question.

4. Prompt and data are assembled.

5. Sent to Gemini API for response.

6. Output displayed in chat UI.

7. Ongoing conversation with preserved context.

?? Technical Components

• Streamlit: Web interface.

• PyPDF2: PDF text extraction.

• Google GenAI Python Client: Gemini API integration.

• Asyncio/Nest-Asyncio: Async operations.

• Dotenv: Secure API key handling.

???? UI Principles

• Simple, clear, accessible.

• Conversational and educational.

• Highlights limitations (e.g., not a diagnostic tool).

6. Conclusion

This research aims to empower patients by simplifying complex health data using generative AI, without crossing into clinical decision-making. While promising, careful attention is needed to ensure:

• Accuracy

• Ethical use

• Privacy

• Human-in-the-loop supervision

It contributes to the growing field of responsible AI in healthcare communication, particularly for non-expert users navigating their medical records.

Conclusion

This project presented a conversational AI system designed to interpret and explain medical-style documents using generative language models. By leveraging tools like Gemini 2.0, ChatGPT, and Claude 3.7, the system demonstrated how AI can improve accessibility to complex health data through natural language interaction. Although tested only with synthetic datasets, the results show promising potential for enhancing health literacy. With further refinement, real-world integration, and expert validation, this approach could become a valuable tool for patient education and support.

References

[1] Bhatt, A., & Vaghela, N. (2024). Med-Bot: An AI-Powered Assistant to Provide Accurate and Reliable Medical Information. arXiv:2411.09648. https://arxiv.org/html/2411.09648v1 [2] Li, Y., et al. (2023). ChatDoctor: A Medical Chat Model Fine-Tuned on a Large Language Model Meta-AI (LLaMA). arXiv:2303.14070. https://arxiv.org/abs/2303.14070 [3] Toma, A., et al. (2023). Clinical Camel: An Open Expert-Level Medical Language Model. arXiv:2305.12031. https://arxiv.org/abs/2305.12031 [4] Zhang, H., et al. (2023). HuatuoGPT, Towards Taming Language Model to Be a Doctor. arXiv:2305.15075. https://arxiv.org/abs/2305.15075 [5] Zhao, L., et al. (2024). CareBot: A Pioneering Full-Process Open-Source Medical Language Model. arXiv:2412.15236. https://arxiv.org/abs/2412.15236 [6] Pap, I.A., & Oniga, S. (2024). eHealth Assistant AI Chatbot Using a Large Language Model to Provide Personalized Answers through Secure Decentralized Communication. Sensors, 24(18), 6140. https://doi.org/10.3390/s24186140 [7] IJISRT. (2024). Leveraging LLM: Implementing an Advanced AI Chatbot for Healthcare. Retrieved from https://ijisrt.com/assets/upload/files/IJISRT24MAY1964.pdf [8] IRJMETS. (2024). Jain, Harsh. “A MEDICAL CHATBOT: YOUR HEALTHCARE ASSISTANCE.” International Research Journal of Modernization in Engineering Technology and Science (2024): n. pag. https://www.irjmets.com/uploadedfiles/paper//issue_6_june_2024/58850/final/fin_irjmets1717770049.pdf [9] (JAMA Network Open, 2024). Huo B, Boyle A, Marfo N, et al. Large Language Models for Chatbot Health Advice Studies: A Systematic Review. JAMA Netw Open. 2025;8(2):e2457879. doi:10.1001/jamanetworkopen.2024.57879 https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2829839 [10] PMC. (2024). Generative AI and Large Language Models in Health Care. Retrieved from https://pmc.ncbi.nlm.nih.gov/articles/PMC10920625Raza, Marium M et al. “Generative AI and large language models in health care: pathways to implementation.” NPJ digital medicine vol. 7,1 62. 7 Mar. 2024, doi:10.1038/s41746-023-00988-4. https://pmc.ncbi.nlm.nih.gov/articles/PMC10920625/

Copyright

Copyright © 2025 Syed Abrar Hassan, Syed Amaanullah, Mohammed Ameer Khan. 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.