An exercise in querying different REST apis, and pulling them together for composing queries to Gemini for hypothesis generation.
Imagine you're looking to think about generating hypotheses: you're studying a disease, you have some protein that's implicated, you sense that there's something worth pursuing. But how do you flesh this out into a larger hypothesis that's (a) plausible and (b) supported by some other evidence? It's 2025, let's try to off-load some of the literature search and curation to LLMs.
- Install
uv
- clone the repo, cd into the repo
- run
uv sync
to re-create thevenv
- configure credentials to use the Google Gemeni API (e.g setting up
GOOGLE_API_KEY
as an environment var) - use the notebook in
src/drive)_hypothesis_generation.ipynb
- module for REST api IO and data transformation
- module for packaging and querying the Gemini API directly
- Minimal set of unit tests to ensure the various components work.
- The Gemeni queries should probably should be mocked, or at least have the response interface better defined so I can avoid hitting the api in tests.