Inspiration
Doctors spend over 15 hours per week on administrative tasks, with a significant portion dedicated to searching through patient histories. This inefficiency leads to burnout and reduces the time available for patient care. We wanted to leverage AI to streamline medical data retrieval, allowing doctors to focus on what matters most—their patients.
What it does
AIMed is an AI-powered medical search engine that enables doctors to instantly retrieve relevant patient history from EMR systems like Epic. Instead of manually sifting through records, doctors can query medical terms, and AIMed will return the most relevant records using vector embeddings—all while ensuring patient privacy and security.
How it would be built
- Frontend: React, Next.js, Tailwind
- Backend: Node.js, FastAPI, Flask, GraphQL
- AI Model: Llama & Ollama for natural language processing
- Database: Elasticsearch for vector-based search
- Security: OAuth 2.0, JWT authentication
- Data Handling: Patient data is anonymized, converted into vector embeddings, and matched with historical records using similarity search
Challenges we ran into
We had started with one idea but ended up switching between 3 or 4.... which took up a lot of time. But in the end, we decided on an idea we were happy with and could go about designing.
What we learned
On the technical side:
- How to optimize Llama for medical NLP and knowledge retrieval
- The importance of federated learning & synthetic data for secure AI training
- Best practices for handling sensitive medical records securely
But other than that we learned to plan, plan, and plan again.
What's next for AIMed
In the future, we would want to create the actual application, as well as integrate:
- Deeper EMR integration with additional systems beyond Epic
- Expanding AI capabilities to suggest diagnoses based on patient history
- Implementing voice-to-text AI for hands-free documentation
- Testing real-world deployments with healthcare providers to refine accuracy
Log in or sign up for Devpost to join the conversation.