-
Medical Assistant giving supported answers using RAG and Gemini
-
Image analysis using RAG and Gemini
Inspiration
We noticed that AI tools can answer surgical questions very confidently, even when they’re wrong. In medicine, that’s dangerous. We wanted to build something that answers using real surgical guidelines and avoids guessing.
What it does
Surgical Tutor is an AI study helper for surgical trainees. It looks up answers from trusted surgical guidelines and only responds if it finds solid support. If it’s unsure, it says so instead of making something up.
How we built it
We used medical text embeddings, a simple knowledge graph, and the Gemini API to generate answers. The system first finds relevant guideline sections, checks them, and then creates a grounded response.
Challenges we ran into
It was tricky to balance being helpful and being safe. If the system is too strict, it refuses too often. If it’s too loose, it risks giving wrong answers. Building the medical knowledge connections was also challenging.
Accomplishments that we're proud of
We built a working system trained on ~447 guideline sections and reduced incorrect or unsupported answers during testing.
What we learned
AI can sound confident even when it’s wrong. Adding structure and checks makes it safer, especially in healthcare.
What's next for Surgical Tutor
We want to add more guidelines, test with more clinicians, and improve how the system explains its answers.
Team Members:
Manglam Srivastav, Abhimanyu Pandey
Log in or sign up for Devpost to join the conversation.