AI tooling for scientific model metadata
I built a tool for Ersilia that helps researchers collect information about machine-learning models. It reads scientific papers, suggests the relevant fields and lets the team check them before publishing.
Fullstack Developer
Ersilia
Feb–Sep 2025
- Context
- Scientific model metadata tooling
- Organization
- Ersilia
- Role
- Fullstack Developer
- Period
- Feb–Sep 2025
About the project
The team was collecting model information through several manual steps. I worked with researchers and developers to bring that work into one application, connected to the GitHub repositories they already used.
My work
- Built the application and backend, including the database for model descriptions and other metadata.
- Added AI-assisted PDF reading so researchers could review suggested values instead of copying everything by hand. The suggestions still needed validation and human review.
- Connected the application to GitHub API and Actions for validation and repository updates.
Built with
Next.js, React and TypeScript for the application; Node.js and PostgreSQL for the backend and data. Vercel AI SDK handles the calls that extract information from papers. GitHub Actions checks and synchronizes the metadata.
Related work
Working on something like this?
I take selected product work and I am open to senior engineering roles. Email is fastest.
Get in touch