INQUIZ
Teachers at prep courses for Azerbaijan's DIM and MIQ exams run quizzes every week, often from paper or photos, and check them by hand: about 1.5–2 minutes per sheet (teacher interviews). For a center with 30–40 classes and 5–6 subjects a week that is roughly 56–120 teacher-hours a week. Teachers get scores, not which topics each student is missing, and students get their results late.
Preliminary score
- Value for the user21.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI25.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing16.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility13.0 / 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 70.0 / 100 on stage.
What the judges wrote
- GPT-6 AstraPreliminary84/100
- Strengths
- Your DIM/MIQ workflow connects teachers’ existing quizzes, student submissions and topic maps to a concrete marking-time problem. Your testing is particularly strong in its transparency: you compare topic tagging with a baseline, quantify extraction and solution coverage, and document a false-positive solution and the resulting rollback. Your measured per-quiz costs, teacher-permitted data and defined two-week pilot make the next adoption step credible.
- To improve
- Your next evaluation should use larger, unseen quizzes from several teachers and report topic accuracy, valid-solution coverage and false approvals separately. Your pilot should measure net teacher minutes, including photo corrections and answer-key setup, rather than only marking time removed. Your arithmetic checker cannot certify chemistry reasoning; add teacher approval or domain-specific validation and distinguish arithmetic checks from correctness verification. Before your pilot, add an explicit extraction-review step and restricted access to student results.
- Claude Opus 5.5Preliminary82/100
- Strengths
- Your problem is grounded in teacher interviews and a concrete workflow: hand-marking DIM/MIQ prep quizzes at 1.5–2 minutes per sheet, with no topic diagnosis. A real chemistry teacher used the app with her students, and the adoption metric (teacher minutes per quiz) is clearly named. Your testing is unusually honest: a held-out topic-tagging test against a keyword baseline (5/5 vs 1/5), per-question photo-extraction metrics, and documented failures. Those failures include 6 of 20 options getting a reason, only 1 useful; 16/16 failed free-tier calls; and a pseudo-solution that passed the arithmetic check. Each one led to a concrete fix, such as hiding unverified reasons, reverting solve-v2 and adding JSON repair. Letting code verify AI solution steps against the teacher's answer key is a thoughtful safeguard rather than blind trust in the model.
- To improve
- Your core accuracy evidence rests on very small samples: 5 questions for topic tagging, 2 pages for photo extraction, and one run each. Re-run on several full quizzes across subjects and report agreement with teacher labels. The time-saving outcome is still a teacher estimate (×1.75 min), so measure actual minutes in the planned 2-week pilot and compare them with the hand-marking baseline. Verified solutions currently cover only arithmetic-heavy questions (chemistry 1–2 of 5 per quiz), so say clearly what students see for non-numeric questions. The pre-built explanation library on the demo quizzes should be visibly separated from live AI output in the demo. Finally, replace the Vite template boilerplate at the top of your README with a short project overview.
The team
Raiq Murtuzəliyev
business
Məhəmməd MirzəyevLead
developer