First and foremost, I do not have any experience in Clinical Trials. My sister, however, does. She has extensive experinece with navigating and running them. According to her, one of the largest problems in this space is just coordination and finding the right people. The incentives, for the most part, are not there. As a first major project after my time at MSFT and learning some things in Comp Bio, I wanted to
- Create something that she could potentially use immediately.
- Play with real world data
- Experiment with searches/patterns and filtering
- Test out the current limitations of what can and can't be done with "Vibe Coding" in order to see what my impression of it is.
This blog is just a running list of my technical thoughts as I developed this-- mostly to give myself (and the agent I'm using) some background into the context, story trials etc. of things as they get done.
Product Direction
A user searchable tool in order to find the "Best" PI based off a given set of contraints.
Constraints
- Try to avoid reinvetning the wheel
- Minimal resource consumption
- Speed to the initial release was the most important constraint
- No manual coding -- everything done via some sort of AI.
Future directions
- Improve the search
- Make sure functionality actually works
- See how long it takes to get a
- Externalize the MeSH tools here to become some sort of reusable API or library.