BasuraBantay: AI eyes for cleaner barangays, from the household bin to the dump truck route

Problem Statement

More than two decades after the Ecological Solid Waste Management Act (RA 9003) of 2000 required segregation at source and an MRF in every barangay, compliance is still uneven. Metro Manila alone generates thousands of tons of garbage daily, and mixed waste still ends up in landfills, rivers, and Manila Bay. The Philippines is often cited as one of the largest sources of ocean plastic. Barangay enforcers cannot inspect every bin. Collection trucks run fixed routes regardless of how full the bins are, wasting fuel. Illegal dumping along creeks and vacant lots goes unreported until it clogs drainage and worsens the floods that come with every typhoon.

Proposed Solution

BasuraBot has three modules. (1) Segregation Checker: a camera at the MRF conveyor or tipping area, or a resident’s phone, photographs a waste bag or bin. A vision model detects items (PET bottles, sachets, food waste, residuals, hazardous items such as batteries) and scores segregation compliance per household or collection point. Residents get feedback by SMS in Filipino. (2) Smart Routing: fill-level estimates from photos or reports, plus the collection schedule, feed a vehicle-routing optimizer that builds daily truck routes to cut kilometers and fuel. (3) Dumping Watch: a model watches existing barangay CCTV feeds and citizen-submitted photos and flags new garbage piles at known hotspots, then dispatches enforcers. A dashboard tracks barangay-level compliance against RA 9003 targets.

Target Market

• Primary users: Barangay ecological solid waste management committees, MRF operators, and city and municipal ENRO offices.
• B2G buyers: LGUs (especially the Metro Manila cities, Cebu, Davao, and Iloilo), MMDA, DENR-EMB, and the National Solid Waste Management Commission.
• B2B buyers: Private haulers, malls and subdivisions, and companies meeting Extended Producer Responsibility (EPR Act, RA 11898) obligations to recover plastic packaging.
• Size: 1,600+ cities and municipalities and 42,000+ barangays. Starts with highly urbanized cities.

How AI Is Used

• Object detection (YOLO trained on waste datasets plus local sachet/PET images) classifies waste and scores segregation. Manual inspection cannot scale to every household.
• Change detection / anomaly detection on CCTV and geotagged photos flags new illegal dump sites automatically, so no one has to watch footage 24/7.
• Forecasting predicts waste volume per route, for example after paydays, fiestas, or typhoons.
• Optimization (vehicle routing with capacity constraints) builds truck routes. Without it, routes stay static and wasteful.
• LLM writes friendly feedback in Filipino or the local language and summarizes compliance reports for barangay captains.

PLDT / Smart / ePLDT / Rokid Angle

• Smart IoT/LTE-M and SMS connect MRF cameras, truck GPS trackers, and resident notifications.
• ePLDT cloud hosts video analytics and the LGU dashboard as a smart-city SaaS. Edge processing reduces bandwidth.
• PLDT Enterprise B2G can bundle it into smart-city and CCTV deals with LGUs.
• Rokid AR glasses give MRF workers and enforcers hands-free item classification and on-site violation logging.
• ESG and EPR: it supports PLDT’s and corporate clients’ sustainability and plastic-recovery reporting.

Demo Scenario (3 minutes for judges)

1. Show a photo of a flooded street in Malabon with clogged drains full of sachets.
2. A teammate places a mixed bag (bottle, banana peel, sachet, battery) under a webcam. BasuraBot boxes each item, scores it “45% segregated,” and warns about the battery as hazardous.
3. The resident gets an SMS: “Salamat po! Pakihiwalay po ang baterya at sachet sa susunod” (“Thank you! Next time, please separate the battery and sachets”).
4. Switch to the routing tab. The AI reroutes three trucks around full bins and a flooded street and shows “−18% km today.”
5. A CCTV clip shows someone dumping bags beside a creek. An alert pops up with the location and timestamp and goes to the barangay tanod.
6. Close with the barangay compliance leaderboard: “From enforcement by guesswork to enforcement by data.”

30-Hour Build Plan

Suggested stack: Ultralytics YOLOv8, OpenCV, Python/FastAPI, Google OR-Tools, OSMnx, React + Leaflet, Gemini or an open LLM, SMS gateway mock. Public data / APIs: TACO (Trash Annotations in Context), TrashNet, Hugging Face waste datasets, OpenStreetMap road network, PSA/NSWMC waste generation statistics, DENR-EMB MRF/sanitary landfill lists, data.gov.ph LGU boundaries.

Risks & Privacy Notes

• Accuracy on local waste: sachets and soiled plastics are hard to classify. Fine-tune on locally collected images and show confidence scores.
• CCTV privacy: under the Data Privacy Act of 2012, faces and plate numbers are blurred automatically. Only the incident (location, time, pile) is logged. Identification is left to authorized officials under due process. Footage retention is limited.
• Shaming risk: household compliance scores stay private and are shown only to the household and the barangay, never published.
• Connectivity at MRFs: run edge inference on a laptop or Jetson and upload summaries only.
• Clause 24 compliance: no real CCTV footage with identifiable people goes into third-party AI tools. Staged or public footage is used for the demo. Dataset licenses (TACO CC BY 4.0) are credited, and AI use is disclosed.

AI Use Disclosure (draft)

We use Ultralytics YOLOv8 (AGPL, open source) for waste and dumping detection, OR-Tools (non-AI optimization) for routing, and Gemini/an open LLM for Filipino feedback messages and report summaries. ChatGPT and GitHub Copilot helped with boilerplate code and pitch copy. Team members reviewed all model outputs, code, and generated text.

Rubric Scoring