Bright Steps
11th-grade students in Azerbaijan preparing for the 2027 DİM English exam. Task 26, a new opinion-paragraph essay graded on 4 official criteria (A–D), is worth 16.7 of the 100 English points. Today a student gets feedback only from a teacher or tutor: it takes days, usually arrives as a single number, and tutors cost money. Students don’t know that grammar (C) and vocabulary (D) are scored on variety, not just correctness, and many memorize template essays, which the rubric scores as zero. They have no fast way to learn why they lost points or to practise their own mistakes.
Preliminary score
- Value for the user21.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI24.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing17.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility12.0 / 15
Data requirements, running costs and a clear next step.
- Originality7.0 / 10
A meaningful difference from common solutions.
A finalist: the live jury scored it 68.0 / 100 on stage.
What the judges wrote
- Claude Opus 5.5Preliminary81/100
- Strengths
- Your problem is unusually specific: DİM 2027 Task 26, its four official criteria, and real student misconceptions such as variety scoring and template essays earning zero. You show a full learning loop rather than a grader alone, running from grading and coaching through practice built from the student's own errors to a rewrite with a before/after card, on a live demo with a test account. Your testing is exemplary for a hackathon. You used a 9-essay golden set with edge cases and repeated runs, compared Sonnet 5 with 5.5 and chose accuracy over speed, and documented concrete failures with fixes: code-computed word counts, dropping hallucinated quotes, structured output, and a blind solver check.
- To improve
- Your golden set of 9 team-graded essays is small. Expand it with essays graded independently by real DİM-trained teachers, and report agreement with them, so the 103/108 figure carries more weight. You have not yet shown any evidence with actual students, such as whether scores improve between first draft and rewrite, which is the outcome metric this track asks for. Prioritise the open photo-reading issue, where the reader silently corrects student errors on blurred images, because it directly corrupts grading. Also add a clear next step covering who pays, how schools or students are reached, and the monthly cost at scale.
- GPT-6 AstraPreliminary83/100
- Strengths
- Your project targets a precise, high-stakes exam task and connects rubric-based grading to mistake-specific practice and a scored rewrite. Your testing is unusually concrete: you compare grading models, disclose ambiguous questions and transcription failures, and describe safeguards introduced after observed defects. Your per-essay cost estimate, deployment instructions and student-data isolation design strengthen feasibility.
- To improve
- Your next validation should use a larger, unseen essay set graded independently by DİM-experienced teachers, separating model repeatability from grading accuracy. Your photo test set should check whether transcription preserves student mistakes and rejects refusal text before grading. For the enterprise track, identify an adopting school or tutoring business and pilot the complete workflow, measuring teacher time saved, student improvement and full-loop cost.
The team
Bayram MammadaliyevLead
developer
Fatima Hasanli
business