DRAGENGrade
In development: pilot schools wantedYour rubric. Your call.
Your evenings back.
DRAGENGrade reads student work against the rubric you already use and suggests a score for each part. Every suggestion comes with a reason that quotes the student's own words, so you can check it in seconds.
Then you decide. You approve, change or throw out every grade. Nothing is final until you say so.
Built to run on the school's own computer. It is being built so grading runs on an AI model on that machine and, on a standard network setup, student work is not sent to a cloud AI.
How it works
Three steps, and the last one is yours.
1. Give it your rubric
The rubric you already grade with: the criteria and what earns each score. It grades against your standard, not one it made up.
2. Give it the students' work
It reads each piece against every line of the rubric and suggests a score for each one, with a short reason.
3. You approve every grade
Each reason quotes what the student actually wrote. Agree and approve it, change the score, or reject it. The grade is yours.
What a suggestion looks like
A score, a reason, and the student's own words.
An illustration of how a suggestion is designed to read. It is not real student work.
A score with no reason is a number you have to trust. A score with the student's own words beside it is one you can check at a glance, and a reason that does not match the work is easy to spot.
It is being built to check its own quotes. If a reason “quotes” words that are not in the student's work, the suggestion is refused. If it still cannot back up a score, that part is marked for the teacher with no points, and the total is shown as incomplete rather than as a number that looks finished.
That is the whole design: the software does the slow first read, and the teacher keeps the judgment. It never grades for you.
The two promises
The teacher decides. Student work stays at school.
Student work is private. So DRAGENGrade is being built around two rules that are part of how it is written, not settings you have to go and find.
- The teacher approves every grade. Every score it suggests waits for the teacher, who approves, changes or rejects it.
- Grading is built to run on the school's computer. The AI model runs on that machine and, on a standard network setup, student work is not sent to a cloud AI. If that model cannot be reached, grading stops and says so, rather than turning to the cloud.
- Nothing goes to a student or a gradebook. The report it writes has an empty column for the teacher's final grade.
- Built for FERPA-sensitive work, with no badges and no “compliant” claims.
Pilot schools
Pilot teachers will shape it, not guesswork.
DRAGENGrade is being built around the way teachers already grade. The next step is building it against real rubrics in real classrooms.
A conversation first
We talk about what you grade, how long it takes, and which rubrics you use most.
Your computers
Because the AI runs on the school's own machines, we look at the computers you have before anything else.
Your say in what it does
Pilot schools shape what DRAGENGrade checks, how its reasons read, and what a teacher sees first.
Want to be a pilot school?
Tell me about your school and what your teachers grade. Please leave student names and work out of the first email.
DRAGENGrade suggests scores for a teacher to review. It does not assign grades or make decisions about students.