EnlightenAI learns from the examples you grade. Each graded sample counts for 20% of training, so 5 graded samples = 100% trained. You can train as you grade, or upload sample gradings up front to calibrate before students submit.
How does the AI get trained?
When you grade a submission, you show EnlightenAI exactly how you score work against your rubric. The AI studies those examples and uses them as context when it generates feedback for the rest of the class.
Training progress is simple math: each graded sample adds 20%. The model reaches 100% at 5 graded samples. You'll see this as a progress indicator on the assignment.
1 graded sample = 20%
2 graded samples = 40%
3 graded samples = 60%
4 graded samples = 80%
5 graded samples = 100%
More samples generally means more consistent feedback, but you can publish and use the AI before reaching 100%.
What is calibration (training up front)?
Calibration means grading example work before the whole class is processed, so the AI is tuned to your standards from the start. You have two paths:
Train as you grade — grade submissions one by one, and the model improves with each one.
Pre-calibrate — upload and grade sample submissions up front in a training assignment, then reuse that trained model on your standard assignments.
Both reach the same place: a model that reflects your rubric and your judgment.
Do I have to reach 100% before grading?
No. The AI can generate feedback at any training level, and you always review and edit before students see it. Reaching 5 graded samples simply gives the model the fullest picture of how you grade.
Still need help? Contact our support team.