AI evaluates. Your rubric decides.
autoG is an independent side project exploring a more dependable way to evaluate open-ended assignments. Instead of asking an AI model to invent its own grading standard, autoG gives it a structured rubric and requires every mark to follow the criteria supplied by the educator.
A grading result you can inspect
The live sandbox accepts an assignment, a hierarchical rubric, optional evaluation context, and a learner submission. It returns criterion-level scores, concise feedback, and exact excerpts from the submission that support each decision. The application validates those excerpts, rejects scores that the rubric does not permit, and calculates category and overall totals in code.
This first version is deliberately focused on the core grading loop. There are no accounts, dashboards, saved submissions, or complex rubric-building tools. The aim is to make the evaluation process easy to try in a browser while keeping the grading engine structured enough to become an API for learning platforms later.
Built for transparent evaluation
The rubric remains the source of truth, the model acts as evaluator, and application logic handles validation and arithmetic. This separation makes results easier to understand and creates a practical foundation for more consistent, auditable assignment feedback.
