Metacare.AI

• 

  I'm Metacare, your helpful AI Assistant

• 

  Ask me a medical related topic.

• 

  I dont answer chats that are not medical questions. I dont like PII/PHI.

• 

  I am helpful, but not a substitute for a doctor or hc professional

• 

  I respect your privacy by not requiring a login or storing any chat information.

• 

  I was built with Docker

• 

  My docker images are deployed on Google Cloud Artifact Registry.

Inspiration

Metacare.AI stems from the ambition to provide individuals with a private medium to discuss medically sensitive subjects, notably topics like sexually transmitted infections (STIs), with a large language model (LLM) such as ChatGPT.

What It Does

Metacare.AI is a cutting-edge platform designed to facilitate private, no-login-required discussions between users and ChatGPT about health-centric subjects. With a focus on ensuring complete anonymity, it promises users a worry-free environment to explore topics like STIs without the fear of personal data exposure.

How It's Built

• Topic Filter: At its heart, Metacare.AI integrates ChatGPT for generative LLM functions. It boasts its own API layer fortified by a BioBert AI model, fine-tuned to sift out spam or off-topic inquiries. Non-medical questions are gracefully deflected, maintaining the platform's core intent.

• Privacy Measures:

  1. No Data Recording & No User Login: Metacare.AI stands out by not requiring any user login, coupled with its foundational design to refrain from recording or logging user interactions, ensuring ultimate privacy.
  2. Topic Filter's Bias Against Personal Information: An inherent bias against personal identifiable information (PII) and patient health information (PHI) ensures data privacy right from the initial interaction.
  3. Anonymization with Microsoft Presidio: An added layer of protection is achieved by integrating with Microsoft Presidio, which meticulously anonymizes any PII within the chat content.

• RAG: The platform seamlessly incorporates a Retrieval Augmented Generation system, empowering users to converse about their medical records, always prioritizing their privacy.

• Fine-Tuning: Through diligent fine-tuning, the platform optimizes its outputs, reducing the possibility of irrelevant or hallucinatory content.

• Deployment: For flexibility and scalability, Metacare.AI is containerized using Docker and then deployed to Google Cloud.

Challenges

Developing AI models to act as platform guardrails presented significant challenges. Curating appropriate datasets and ensuring continuous training and testing were particularly demanding.

Accomplishments

The team takes pride in the robust and efficient guardrails encapsulating ChatGPT. This design not only amplifies the utility of the LLM but also remains steadfast in preserving user anonymity.

Future Prospects

Plans for Metacare.AI encompass open-sourcing its foundational code and model training data for public examination and validation. In the pipeline are features to enhance user experience, such as more seamless integration of medical record information.

Built With

• bert
• gcloud
• javascript
• node.js
• python
• pytorch
• react
• transformers