Inspiration
Navigating healthcare can be overwhelming. Many patients struggle to accurately recall or articulate their symptoms, timelines, and medical history during doctor visits. This often leads to miscommunication, incomplete diagnoses, and suboptimal care. We wanted to build a solution that empowers patients to take control of their health records while also making life easier for healthcare providers.
What it does
MediMate is an AI-powered health companion utilizing the Google Gemini API, designed to streamline communication between patients and doctors. It consists of four key components:
Medical & Family History Input: A dedicated space for users to record their prior medical conditions, treatments, allergies, and family health history.
Daily Health Logs: A simple, intuitive journal for users to log daily symptoms, pain levels, concerns, and overall well-being.
Your Health Companion (AI Chatbot): A conversational assistant that allows users to ask health-related questions. It references the user's history and logs to provide evidence-based responses without diagnosing, acting instead as a easily accessible health consultation resource.
Summarized Health Reports: The AI can generate a comprehensive report focused on a user-specified health issue, compiling relevant calendar logs based on reported symptoms for easier communication with your healthcare provider.
How we built it
Frontend: Built using JavaScript and React for a friendly user experience. Backend: Uses JavaScript, Google Gemini, and Firebase. Gemini API was used for creating the AI chatbot and report generator. We implemented advanced Chain of Thought (CoT) and Reflection on Thought (RoT) prompt engineering techniques alongside previously established literature to fine-tune the AI’s behaviour and ensure safe, informative outputs. Firebase was used for secure user authentication and to store medical history, logs, and personal data efficiently.
Challenges we ran into
Seamlessly integrating the four components so that they work together as a unified system and allow data retrieval between modules. We also had difficulty designing the authentication system to allow users to revisit and update their health information securely. Lastly, creating prompts that ensure the AI assistant is helpful but stays within its boundaries to avoid medical diagnosis was challenging.
Accomplishments that we're proud of
We successfully built a fully functional tool that not only logs user health data but can also analyze it for trends and patterns. Our research-backed prompting strategies were a staple in the creation of an AI companion that feels personal yet safe, utilizing the user's actual data without overstepping into medical diagnoses. Additionally, our prompts play a crucial role in the generated health reports. We were able to expertly use Google Gemini to filter through irrelevant calendar logs and pinpoint only the most relevant ones for the report. We stand at the forefront of minimizing the communication issues in patient-doctor interactions.
What we learned
Working on this project reinforced the power and nuance of prompt engineering, particularly how small changes in phrasing can significantly impact the behaviour of an AI model. We incorporated various techniques, such as CoT and RoT, that have been developed to effectively engineer prompts, guiding our own AI companion to respond with friendly yet informative answers.
What's next for MediMate
We believe that MediMate paves the way for healthcare documentation. Further expansion of the AI companion to auto-write logs for the user could improve the usability of our tool. Additionally, incorporating a speech-to-text function would increase accessibility. Daily habit tracking, including lifestyle factors like diet, exercise, and sleep, would provide a more holistic view of one's health. Expanding to a Doctor-side Interface to create a professional dashboard for doctors to view generated patient summaries ahead of appointments could allow doctors to have a deeper understanding of their patients' everyday lifestyles, which could improve decision-making and reduce administrative busywork.
Log in or sign up for Devpost to join the conversation.