ProtocolGraph

Inspiration

We wanted to build something about biotechnology and AI, and were inspired by the starter material provided by Regeneron. We were looking through some sample clinical trial protocols, and did not appreciate how lengthy, jargon-filled, and verbose the documents were.

What it does

ProtocolGraph uses the Gemini API and Python's NetworkX library to model PDF documents of clinical trial protocols as concept maps involving relationships between entities present in the document. One can also simulate scenarios by adjusting different variables that alter the concept map.

How we built it

We used React and Cytoscape.js for the frontend, Python's FastAPI and NetworkX for the backend, and SQLite for the database. We made the Gemini API pass a JSON object into two validators, one for malformed JSON objects and another for logical consistency, and stored a serialized JSON representing the NetworkX graph inside a SQLite database for future retrieval and scenario simulation.

Challenges we ran into

Not much actually, it was pretty smooth sailing

Accomplishments that we're proud of

Developing a coherent system design under constraints

What we learned

Proper steps of the SDLC and prototyping using system design and whiteboarding

Built With

• cytoscape
• fastapi
• gemini
• networkx
• python
• sqlite
• typescript