Problem Statement
Teachers spend nights writing quizzes by hand, and photocopy budgets are thin. Projectors and internet are rare; pupils share tattered books. AI quiz tools need uploads of copyrighted pages to clouds teachers don’t trust and connections schools don’t have.
Proposed Solution
Photograph the lesson (book page, board, handwritten notes). On-device OCR extracts text; a small quiz-writer drafts 10 items (multiple choice + 2 short answer) with an answer key and difficulty tags aligned to the lesson. All on-device; no page photo ever uploads. Export as large-print handout or dictation script for zero-paper classes.
Target Market
- Primary users: Public school teachers, ALS facilitators, mission-school volunteers.
- Deployers: Department heads sharing quiz banks via SD card.
How AI Is Used
- OCR + cleanup: Handles mixed print/handwriting and two-column pages.
- Quiz-writer LM: Generates distractors that catch common misconceptions, not trivia.
- Why AI: Manual authoring is the bottleneck; templates produce shallow recall items.
Independent Deployment Angle
- No cloud, no LMS subscription, no venue, no headset. One teacher phone suffices.
- Quiz banks move on SD cards teachers already pass around.
- Respects book copyright: photos stay on-device, only teacher-authored items export.
Demo Scenario (3 minutes for judges)
- Airplane mode. Snap a mock lesson paragraph the team wrote.
- 10-item quiz + key appears in seconds with easy/medium tags.
- Show dictation mode: app reads Q1 aloud for zero-paper recitation.
- Export large-print handout preview.
- Close: “Isang litrato, sampung tanong.”
30-Hour Build Plan
| Hours | Task | Role |
|---|---|---|
| 0–3 | Write mock lessons, define quiz spec | All |
| 3–12 | OCR + quiz-writer prompts, distractor quality | AI lead |
| 3–10 | Camera + bank + export | Backend |
| 10–20 | Teacher UI, print CSS | Frontend |
| 20–26 | Teacher review pass, misconception check | All |
| 26–30 | Pitch, disclosure | PM / Pitch |
Suggested stack: Flutter + on-device OCR, quantized quiz-writer TinyLM, SQLite bank. Public data / APIs: Team-written mock lessons; no external API.
Risks & Privacy Notes
- Copyright: Demo lessons are team-written; app warns not to distribute book photos. Only generated items export.
- Quality: Teacher approves every item; key shows source line. No auto-grading of pupils in v1.
- Privacy: Pupil names never entered; no personal data into AI tools.
AI Use Disclosure (draft)
Uses on-device OCR and a small LM for quiz generation. AI coding assistants helped with export templates. All demo items were reviewed by the team for correctness.
Rubric Scoring
| Criterion | Weight | Score (1–5) | Why |
|---|---|---|---|
| Innovation & creativity | 25% | 4 | Offline photo-to-quiz for low-resource rooms is strong |
| Problem–solution fit | 25% | 4 | Saves nightly prep time |
| AI application & value | 20% | 4 | Generation + distractors need LM |
| Feasibility | 15% | 4 | Very demoable |
| Impact & scalability | 15% | 4 | Every teacher benefits |
| Weighted total | 4.10 |