Problem Statement
After habagat floods, DOLE deployed TUPAD emergency jobs to tens of thousands across Bataan, Bulacan, Pampanga, Tarlac — but workers ask: bakit wala ako, kulang ang sahod, sino yung pangalang di taga-dito? Payout sheets are posted as blurry photos; ghost names thrive in confusion. LGU staff answer the same questions 200x a day.
Proposed Solution
Type name + barangay (or SMS with no data): TUPADWatch checks public payout postings and your proof of work, explains status (“Nasa batch 3 ka, payout Aug 12-15, dalhin ang ID”), and flags anomalies (duplicate names, non-resident addresses, double payouts) for audit. If denied, it drafts a polite appeal letter with requirements checklist. All queries logged for transparency, no fixers.
Target Market
- Primary users: Habagat-affected workers, solo parents, fisherfolk in Central Luzon and NCR.
- B2G: PESO offices, barangay secretaries, DOLE regional teams for triage.
How AI Is Used
- List reader: OCR + LLM reads blurry payout photos into searchable rows.
- Anomaly scorer: Flags duplicates and out-of-barangay patterns.
- Appeal drafter: Writes Tagalog appeal with correct attachments.
PLDT / Smart / ePLDT / Rokid Angle
- Smart: Free SMS status checks for no-load workers.
- ePLDT: Hosts audit logs in-country for DOLE/LGU.
- PLDT Enterprise: PESO bundle with connectivity.
Demo Scenario (3 minutes for judges)
- Show mock payout photo; OCR turns it into searchable list live.
- Query “Juan Dela Cruz, Hagonoy” — status + payout window appears.
- Flag view: 3 duplicate names highlighted for audit.
- Appeal draft generates for a denied mock worker.
- Close: “Walang palakasan, may paliwanag.”
30-Hour Build Plan
| Hours | Task | Role |
|---|---|---|
| 0–3 | Mock payout sheets + appeal rules | All |
| 3–12 | OCR + explainer + anomaly scorer | AI lead |
| 3–10 | Search + SMS + logging | Backend |
| 10–20 | Worker-first UI | Frontend |
| 20–26 | Edge-case testing | All |
| 26–30 | Pitch, disclosure | PM / Pitch |
Suggested stack: Next.js, FastAPI, LLM list-explainer, anomaly detector, Postgres, SMS query. Public data / APIs: DOLE press releases, public payout postings (mock for demo).
Risks & Privacy Notes
- No shaming: Flags are audit leads, never public accusations. Demo names fictional; no real beneficiary data into AI tools.
- Privacy: Queries minimized, logs aggregated; complies with Data Privacy Act.
- IP: Mock sheets team-made; no real lists scraped.
AI Use Disclosure (draft)
Uses OCR plus LLM explanations and anomaly scoring. Coding assistants helped with UI. All demo verdicts human-checked.
Rubric Scoring
| Criterion | Weight | Score (1–5) | Why |
|---|---|---|---|
| Innovation & creativity | 25% | 4 | Fair-aid explainer with ghost detection is timely |
| Problem–solution fit | 25% | 5 | Post-habagat confusion is real |
| AI application & value | 20% | 4 | OCR + reasoning essential |
| Feasibility | 15% | 4 | Very buildable |
| Impact & scalability | 15% | 4 | Nationwide TUPAD use |
| Weighted total | 4.40 |