Open science · Applied AI
Pipette.
Every morning Pipette reads every new paper on arXiv, bioRxiv, medRxiv and 58 leading journals, and publishes a short edition of the ones worth knowing about. It never rewrites science: for each paper it quotes the sentence of the abstract that states the main result. Free, no ads, no accounts, and open source.
- Result
- Two thousand papers read a day, free and ad-free.
- Time
- Launched in September 2026
- Built with
- Next.js, Python, AWS Lambda, S3, Jev (TypeSafe), Vercel
How the project went
Why they called us
Thousands of new papers appear every day, and AI is about to multiply that number. Existing tools either search for something specific or summarize with AI, putting words in scientists' mouths.
The problem they had
Finding what matters each day without reading two thousand abstracts, without one field drowning out the rest, and without a machine inventing what a paper says.
How we solved it
We use Jev, a decision model that cannot generate text: it only chooses among options and reports calibrated probabilities. It answers ten questions about every paper, a public formula picks the edition, and readers always see the authors' words.
Technical approach
A daily Python AWS Lambda reads arXiv, bioRxiv, medRxiv and OpenAlex, asks Jev and publishes open JSON to S3. The site is Next.js on Vercel, in English and Spanish, with field and topic pages, sitemaps listing every paper, structured data and IndexNow pings.
Impact
About two thousand papers read every weekday, for roughly 17 US cents of model usage. The data is published under CC0 and the code is open on GitHub.
How long it took
Launched in September 2026.
What we needed from the client
None: it is Ferced's own product, built and kept free for everyone who reads science.
Process friction
Medium: ranking science from every field with honest criteria, in two languages and without generating text, needed clear editorial rules as much as code.