MedNav

Inspiration

I often find myself using ChatGPT to discuss health concerns that I believe are not severe enough to see a doctor immediately. Often times, when speaking back and forth with Chat, I wonder what would happen if I ignored a potentially serious symptom. From a study we found on PubMed, "Why do People Avoid Medical Care? A Qualitative Study Using National Data" (Taber et al., 2014), we found that the hesitance to see a doctor immediately is common due to barriers such as high costs and lack of time. In hopes of improving healthcare accessibility by providing a more trustworthy chatbot and a map function to streamline selecting healthcare providers, we are introducing MedNav, a medical guidance tool featuring an AI healthcare navigation assistant.

What it does

MedNav features a trustworthy chatbot, Medly, where users can enter in their symptoms. The chatbot, designed to focus on reliability and trustworthiness by retrieving from grounded sources (Mayo Clinic, etc.), will provide various explanations, prompt the user with relevant questions, give estimated severity and recommended next steps, feed into the map feature where the user can search and book appointments, and provide key notes to bring to the appointment. Additionally, there are features such as notifying emergency services to improve notice for healthcare employees and having the option to save a summary of chats for future reference.

How we built it

MedNav integrates structured reasoning with real-world care navigation. We designed the backend using FastAPI and Pydantic, centered around an orchestrator that manages the full conversation flow. We broke the system into stages across Intake, Retrieval, Moderator, and Recommendation agents, to infer the users intent, gather and reason over medically grounded information, and provide the user with recommended next steps and the means to execute them. Our frontend was designed through Figma, and built using React (Vite + TypeScript).

Challenges we ran into

It was difficult to maintain a clean schema for reasoning while producing fluid, user-friendly chat responses. Additionally, access to reliable medical knowledge sources generally requires forms and applications due to privacy concerns, limiting our ability to enhance our model's capabilities.

Accomplishments that we're proud of

We built a fully conversational end-to-end care navigation system that emphasizes safety, connecting AI reasoning with real-world action. Our system provides explainable recommendations, such as possible causes, supporting reasoning, trusted medical sources, and suggested providers.

What we learned

We learned a significant amount about full-stack development and agentic AI systems, as well as general concerns relating to accessibility and costs of healthcare.