GitHub Repository
Health-Universe/diabetes-llm
Status
Ready
Created On
Updated On
Diabetes Companion is an innovative, AI-powered application designed to support individuals managing diabetes. Leveraging the advanced capabilities of GPT-4, the app provides informative, empathetic, and contextually relevant guidance to users with Type 1, Type 2, or gestational diabetes. It aims to empower users with knowledge and support, enhancing their ability to understand and manage their condition effectively. It uses the Philter library to deidentify the user's inputs and remove PII/PHI from it.

More details

Use Cases Limitations Evidence Owner's Insight
  1. Newly Diagnosed Patients: For individuals recently diagnosed with any type of diabetes, the app provides foundational knowledge about their specific condition, helping them understand the basics of diabetes management.

  2. Long-term Diabetes Management: Offers ongoing support and updated information for those who have been managing diabetes for an extended period, focusing on new research, advanced management strategies, and lifestyle adaptation.

  3. Diet and Exercise Guidance: Assists users in planning and maintaining a diabetes-friendly diet and exercise routine, offering suggestions based on general guidelines.

  4. Blood Sugar Monitoring Assistance: Provides guidance on how to effectively monitor blood sugar levels, interpret readings, and understand when to seek medical assistance.

  5. Emotional Support and Community Engagement: Connects users with support groups and forums, offering a platform to share experiences and receive peer support.

  1. No Personalized Medical Advice: The app does not offer personalized medical advice or treatment plans, which should always be sought from healthcare professionals.

  2. General Guidelines: Recommendations on diet, exercise, and management are based on general guidelines and may not suit every individual's specific needs.

  3. Dependence on User Input: The effectiveness of the app depends on the accuracy and completeness of the information provided by the user.

  4. Technology Constraints: Users with limited access to technology or insufficient technological skills may find it challenging to utilize the app effectively.

  5. Data Privacy: When the authors of the paper "Protected Health Information filter (Philter): accurately and securely de-identifying free-text clinical notes" evaluated the Philter library, it removed 99.4% of PHI in the clinical notes. While impressive, it is not perfect and can sometimes miss edge cases. Note that OpenAI, the service that powers our AI, doesn't use your conversations to make their AI smarter. They only keep the information for 30 days to check for any misuse, and then it's deleted permanently (unless legally required otherwise).

AI Model and Training: The Diabetes Companion app is powered by the GPT-4 model, which has been trained on a diverse array of texts, including peer-reviewed articles and clinical guidelines, up to April 2023. This training foundation ensures that the app's responses are grounded in a broad spectrum of reliable sources. Additionally, the model has undergone extensive Reinforcement Learning from Human Feedback (RLHF), focusing on enhancing the relevance, accuracy, and safety of its responses, particularly in the context of diabetes care.

Specialized System Prompt: To further refine the app’s capabilities, we have developed a specialized system prompt tailored to the needs of individuals managing diabetes. This prompt steers the AI to provide more precise and contextually appropriate information, aligning with the latest best practices in diabetes management.

I'm Venkata Chengalvala, an AI consultant for Health Universe. I port high-quality peer-reviewed AI models helpful to clinicians, patients, and/or researchers to Health Universe's platform. I have a Bachelor of Science in Molecular, Cellular, and Developmental Biology (MCDB) and Computer Science from the University of Michigan-Ann Arbor. During my education, I engaged in medical research, co-authoring a literature review on tumor-derived exosomes, cited by 30+ people so far: https://www.sciencedirect.com/science/article/pii/S2211383521001398.

I created the app to provide accessible, accurate, and supportive resources for individuals managing diabetes, hence empowering users with knowledge and support and enhancing their ability to manage their condition effectively.

Your insights and feedback on this app are greatly appreciated. If you have any thoughts, experiences, or feedback, please leave a comment below.

Stable

Warning: App may appear to work well but has not been peer reviewed. Not intended for clinical use. Use with caution.


  • Favorites: 0
  • Executions: 9

  • Natural Language Processing

Owner

Venkata Chengalvala

Member since