Problem Statement
Palengke prices swing daily and rumors inflate them (“₱400 na daw ang sibuyas!”). DA price watches are weekly and Manila-centric. Inside concrete palengkes signal dies, so buyers can’t check and vendors can’t prove fairness. Both sides lose trust.
Proposed Solution
Snap the price board; on-device OCR extracts items and prices. Nearby phones confirm via Bluetooth (“Tama, ₱120 ang kamatis kay Aling Rosa”). Confirmed prices form the day’s board; weekly index shows trends per stall section. No central server to game; market association holds a read-only export for transparency.
Target Market
- Primary users: Palengke vendors, buyers, market masters.
- Deployers: Market associations, consumer groups, agri coops tracking farmgate-to-retail gaps.
How AI Is Used
- On-device OCR + cleanup: Reads handwritten boards and normalizes “kamatis 120/kilo” variants.
- Agreement scoring: Embedding match merges duplicates; confirmation quorum promotes a price.
- Price-note writer: Small LM writes “Bumaba ₱20 ang ampalaya vs kahapon” summaries.
Independent Deployment Angle
- No cloud price server, no carrier deal, no venue, no headset. Market owns its data.
- Syncs inside dead-signal halls via Bluetooth; one phone with signal can publish the public board.
- Open schema any market can copy.
Demo Scenario (3 minutes for judges)
- Snap a mock price board photo; OCR fills items and prices offline.
- Second phone confirms two prices; board gains “confirmed” badges.
- Show weekly trend for ampalaya from 5 days of mock entries.
- Airplane-mode sync between two phones inside the demo.
- Close: “Presyong pinagkasunduan, hindi haka-haka.”
30-Hour Build Plan
| Hours | Task | Role |
|---|---|---|
| 0–3 | Photograph mock boards, define item list | All |
| 3–12 | OCR + normalization, Tagalog item variants | AI lead |
| 3–10 | Bluetooth confirm + quorum logic | Backend |
| 10–20 | Board + trend UI | Frontend |
| 20–26 | Market-noise test, handwriting edge cases | All |
| 26–30 | Pitch, disclosure | PM / Pitch |
Suggested stack: Flutter + on-device OCR, Bluetooth sync queue, SQLite. Public data / APIs: DA price watch (reference only), team-made mock boards.
Risks & Privacy Notes
- Price gaming: Quorum + stall-section scoping limits manipulation; market master can flag abuse.
- Privacy: No personal data; stall names only with vendor consent. Demo stalls fictional.
- IP: Item lists team-built; OCR libs licensed and credited.
AI Use Disclosure (draft)
Uses on-device OCR plus a small LM for normalization and summaries. AI coding assistants helped with camera code. All demo boards are synthetic and human-checked.
Rubric Scoring
| Criterion | Weight | Score (1–5) | Why |
|---|---|---|---|
| Innovation & creativity | 25% | 4 | Peer-confirmed offline price board is fresh |
| Problem–solution fit | 25% | 4 | Trust + dead-signal fit |
| AI application & value | 20% | 4 | OCR essential for boards |
| Feasibility | 15% | 4 | Demoable with mock boards |
| Impact & scalability | 15% | 4 | Every palengke can copy |
| Weighted total | 4.10 |