Gradus reads a resume against a job posting and answers Keep, Explore or Skip, with one line of why.
Best Use of Adaption, 2026
How it works
Some postings rule you out before skill matters. Gradus checks those first.
Nothing rules you out, and you meet what it asks for.
Worth a look, but not a clean fit.
A hard requirement rules you out.
Adaption reads each posting into a requirement card. Rules check it against the profile before the model runs.
Alex is on an F-1 visa, so this one is Skip. As a US citizen, Alex gets a Keep at 75%.
A 4B model trained with AutoScientist reads the resume and the whole posting, then says why.
The candidate is an enrolled CS undergrad graduating in 2028 with Java, Python and React experience, and the internship is in Beaverton, OR, where they will relocate
From Alex’s live run, as the model wrote it.
Built with Adaption
Adaptive Data reads the postings and writes the resumes. AutoScientist trains the model.
Real postings, read into requirement cards.
Test resumes written from candidate cards, each checked by code.
One AutoScientist iteration trained the model.
Results
One sealed test, scored once. These are its 513 rows where the resume decides, before and after training.
Accuracy where the resume decides
Macro-F1
Deal-breakers caught
+21.7 over untrained+19.2 over rules alone
Run B clears the pass bar set before the test was built.
Live demo
Add a resume. Gradus sorts about 150 real US tech internships for you.
Checking the model
After the rules, the model scores this many postings live. The rest are listed as not scored.
The live model is paused until next month so the demo stays free to run. You can still see it work:
These were over the live limit. Raise the number under Scoring options to see them.