GrabScore

Hand back a score, a grammar list, and a model essay, and the student stares at it once and forgets it. Under exam pressure, almost none of that comes back. GrabScore goes a layer deeper: it fixes the student’s own argument and phrasing, not somebody else’s essay. Feedback is shaped so it can be reused next time, not admired and shelved.

It starts before the writing does — at the prompt and the outline. The system checks whether the angle actually answers the question, whether the logic links up, and whether ideas connect before the student commits to a full draft. After writing, it reviews each sentence across TR, CC, LR, and GRA with concrete rewrites that keep the student’s reasoning intact.

Scoring and deductions track the official band descriptors line by line, with input from examiner-side review. The point is to keep feedback tied to the same rubric the test is graded on — so it sounds like exam language, not motivational coaching.

Each session can spin out targeted vocabulary recall and translation drills based on the weak points it just flagged, plus a PDF report for offline review. One correction turns into a small study cycle instead of a one-shot transaction.

I built and run the whole thing — product, model workflow, frontend, backend, payments, day-to-day ops. Solo project.

Live with four-dimension scoring. Ongoing work is the long tail: refining Academic vs. General Training, Task 1 chart prompts, common Task 2 question types, and the edges where the rubric gets fiddly.