All productsEducation and training
AutoGrader
AI grading and feedback for programming assignments
An LLM judge that grades student code against a rubric, labels the mistakes, and writes feedback at the level the teacher picks, without giving the solution away.
A 7.6-billion-parameter model, fine-tuned and run on a 16 GB laptop with no GPU.
Screens
What it does
-
Rubric grading
Six component marks and a total out of 10, checked against instructor grades with quadratic weighted kappa.
-
Error labels
Each submission gets labels from a ten-category error taxonomy, measured with macro-F1.
-
Feedback that teaches
Feedback in Vietnamese at the level the teacher asks for. At the hint levels it never includes fixed code or a worked solution.
-
Runs on your own machine
An open-weight 7.6-billion-parameter model, plus a LoRA fine-tune trained on a laptop, so student work never leaves it.
-
Reproducible results
One command reruns the whole system, and every reported number comes with the conditions that produced it.
Want something like this for your business?
Tell us about the process or product you have in mind. We reply by email with questions or a time to talk.
Talk to us