Clara — AI Healthcare Navigation Agent

Clara turns confusing medical paperwork into clear next steps, organizing medications, follow-ups, and care tasks so patients know what happened, what comes next, and what still needs attention.

Comment

Inspiration

Clara started from something personal.

My mom had medical paperwork that was hard to follow, and I wanted a simple way to break it down for her so she could actually understand what it meant, what she needed to do next, and what she still needed to follow up on.

That made me think about how many people leave a doctor’s office or hospital with pages of discharge instructions, medications, lab orders, follow-ups, and medical terms they may not fully understand.

The appointment may be over, but the work is not.

I wanted to build something that could help a person look at their medical paperwork and clearly understand:

What happened? What do I need to do next? What am I still waiting on? That became Clara.

My goal was never to build an AI that tries to be a doctor. Clara is designed to help people understand and organize the information their healthcare team has already given them.

What it does

Clara can review multiple medical documents together and turn them into a more organized care experience.

It can:

explain documented medical information in plain language organize medications and follow-up instructions create a structured care plan identify pending care tasks allow users to mark tasks as completed retrieve medication-specific information preserve care-plan information across sessions using Firestore

One of the features I am most proud of is persistence. If a user completes a task and later refreshes or starts a new session, Clara can retrieve the updated care plan instead of starting over.

How I built it

I built Clara as a full-stack AI agent application using Google technologies.

The frontend was built with React and Vite and is deployed using Firebase Hosting.

The backend uses FastAPI and is deployed on Google Cloud Run.

For the agent itself, I used the Google Agent Development Kit (ADK) to handle agent orchestration and tool use.

Gemini through Vertex AI handles document understanding and reasoning, while Google Cloud Firestore stores the structured care-plan state.

The overall flow is: User ↓ Firebase Hosting ↓ React + Vite ↓ Cloud Run ↓ FastAPI ↓ Google ADK ↓ Gemini via Vertex AI ↓ Firestore

Challenges I ran into

This project definitely challenged me, but I welcomed that.

I am still new to building full-stack AI applications, and one of my goals is to grow into an AI Engineer. Because of that, I treated every issue I ran into as part of the learning process instead of something to avoid.

I had to work through authentication issues, Google Cloud permissions, service accounts, Vertex AI configuration, Firestore persistence, CORS problems, local development ports, Cloud Run deployment, Firebase Hosting, and connecting the frontend to the deployed backend.

There were moments when the frontend worked but the backend did not, and other moments when the backend worked locally but needed additional configuration in the cloud.

Those challenges actually became one of the most valuable parts of the project because they forced me to understand how the pieces fit together.

One of my biggest lessons was realizing that building an AI agent is much more than connecting a model to a chat box.

The model can reason, but the application still needs tools, state, storage, cloud infrastructure, authentication, safety rules, and a clear workflow around that intelligence.

Accomplishments that I'm proud of

Honestly, I'm proud that I was able to complete this project at all. I do not have a career in engineering or development ,even though that is my dream. I decided to partake in this Hackathon to gain experience and confidence. Often I feel like I cannot be successful as an engineer because I am not as inept in coding. In this sea of participants my project may be all but underwhelming but knowing that I put my all into it and what it meant to build something that could assist my mom makes me Happy! Thank you for this opportunity to reach for my dreams. I'm proud to put this project in front of you.

What we learned

Building Clara gave me a much stronger understanding of what it actually takes to build and deploy an agentic AI application.

I learned how to:

-orchestrate an agent with Google ADK -use Gemini through Vertex AI -create and call tools from an agent -maintain persistent state with Firestore -deploy a FastAPI backend to Cloud Run -deploy a React frontend with Firebase Hosting -connect frontend, backend, AI reasoning, tools, and cloud storage into one working system -design guardrails around a healthcare-related AI use case -troubleshoot a real application across local development and production

As someone aspiring to become an AI Engineer, Clara gave me the chance to move beyond simply learning concepts and actually build something end to end.

I also learned that useful AI does not always have to be the most complicated AI.

Clara is valuable because it takes information people already have and makes it easier to understand, organize, and act on.

What's next for Clara — AI Healthcare Navigation Agent

Clara is currently a working prototype, but I see a much larger future for it.

One of the biggest next steps would be integrating Clara with patient portals such as MyChart, so users could securely bring in visit summaries, discharge instructions, medications, lab information, and follow-up details without manually uploading every document.

Future versions could also include:

-secure patient accounts -MyChart or other patient-portal integrations -appointment reminders -medication schedules -accessibility features -more personalized care navigation -smarter follow-up tracking across multiple visits

The long-term vision is for Clara to become a secure bridge between the medical information patients receive and the actions they need to take afterward.

Your care, made clear.

Built With

Share this project:

Updates

Submission history