Inspiration

Riding through urban and suburban corridors across Indonesia including the industrial arteries around Cikarang reveals a stark, ubiquitous reality: roads plagued by hazardous potholes that remain neglected for months, or even years.

Official data from Statistics Indonesia (BPS) and the Ministry of Public Works and Public Housing (PUPR) confirms that over 87,000 kilometers of regional roads are classified as damaged or severely damaged. The human cost is devastating: the Indonesian National Police Traffic Corps (Korlantas Polri) recorded at least 1,838 traffic accidents caused directly by road defects within a single year.

Behind this physical deterioration lies a deeper governance crisis: fictitious maintenance invoicing (SPJ Fiktif). According to Indonesia Corruption Watch (ICW), public procurement and infrastructure maintenance consistently lead regional corruption cases. Municipal budgets are routinely drawn down for phantom resurfacing projects, while physical roads remain untouched.

Conventional centralized municipal complaint apps fail because internal administrators hold root database privileges—allowing them to quietly mark tickets as "Resolved" or delete citizen submissions to satisfy bureaucratic key performance indicators (KPIs). We asked ourselves: What if public complaints could never be erased, and what if municipal contractors could only claim project completion by submitting cryptographic, AI-verified proof of physical repair? That inquiry inspired CivicPulse.

What it does

CivicPulse is an AI-audited, Web3-anchored public infrastructure verification platform that transforms passive citizen grievances into a tamper-proof Proof of Physical Repair Protocol:

Frictionless Geo-Reporting: A citizen captures a photo of a road defect via a mobile web browser. The application automatically extracts device coordinates using the HTML5 Geolocation API.

AI Vision Gatekeeper: Before entering the public ledger, the image is parsed by Google Gemini 1.5 Flash Vision. The model validates whether the photo contains genuine road asphalt damage, assigns a Pothole Severity Score, and rejects invalid submissions (such as indoor scenes or downloaded stock photos) within two seconds.

Immutable Attestation on Base Sepolia: Validated reports are committed to the decentralized ledger via the Ethereum Attestation Service (EAS) on Base Sepolia Testnet. Once registered, the record is immutable: no regional administrator can delete or alter the entry.

AI Before-After Spatial Matching: When a contractor claims resurfacing work is complete, they must upload an inspection photo at the corresponding location. Our AI spatial-matching engine compares static background landmarks (such as utility poles, curbs, and adjacent building facades) to ensure geographic consistency, while confirming complete asphalt resurfacing.

Deterministic On-Chain Resolution: Once the AI confirms physical execution, the smart contract transitions the report state to RESOLVED, generating an open, real-time audit trail for taxpayers, civil watchdogs, and supreme audit institutions (such as BPK and KPK).

Challenges we ran into

Spatial Landmark Verification vs. Angle Variance: Contractors rarely take post-repair photos from the exact same centimeter or focal angle as the citizen's initial report. Early prompts caused the AI to reject valid repairs due to perspective shifts. We resolved this by engineering our multimodal prompt to isolate invariant structural anchors (e.g., utility pole positions, building rooflines, curb geometry) rather than relying on strict pixel-by-pixel comparisons.

Eliminating Web3 UX Friction: Forcing everyday commuters to connect a Web3 wallet, sign transactions, or pay gas fees would destroy user adoption. We overcame this by abstracting the blockchain layer completely through a backend relayer model, delivering a standard Web2 user experience backed by Web3 immutability.

Latency Management in Serverless Environments: Combining client-side image compression, AI multimodal inference, and on-chain EAS attestation within a single request risked edge function timeouts. We optimized this by designing an optimistic UI update with asynchronous on-chain minting.

Accomplishments that we're proud of

100% Free-Tier & Public-Good Architecture ($0 Stack): We proved that scalable, corruption-resistant civic technology does not require million-dollar municipal budgets. CivicPulse runs entirely on free, open-source, and public-good infrastructure (Vercel, Base Sepolia, Supabase, and EAS).

True Closed-Loop Implementation: Unlike standard hackathon prototypes that stop at "reporting an issue," CivicPulse enforces the complete lifecycle—from initial citizen alert, to AI fraud screening, to contractor proof-of-work, and finally on-chain status resolution.

Live Deployment Without Mockups: Delivered a fully functional Minimum Viable Product (MVP) running live on the Base Sepolia Testnet and public cloud, strictly adhering to competition integrity guidelines.

What we learned

Attestations as the Natural Web3 Primitive: We discovered that public goods and governance solutions rarely need speculative financial tokens. Standardized on-chain attestations (via EAS) provide the ideal neutral primitive for non-financial trust.

Multimodal AI as an Objective Auditor: Large Multimodal Models (LMMs) excel at spatial reasoning and comparative physical analysis when guided by structured JSON schema constraints.

The Speed of Modern Vibe Coding: By combining Next.js 14, shadcn/ui, and specialized AI coding workflows, our team converted a complex system architecture into clean, deployed software within the hackathon marathon constraints.

What's next for Civicplus

Edge Device & Dashcam Integration (DePIN): Collaborating with ride-hailing fleets (ojek online) and logistics couriers to run lightweight vision models on mounted smartphones, enabling automated, passive road defect mapping as drivers complete daily delivery routes.

Zero-Knowledge Proximity Proofs (ZK-SNARKs): Implementing cryptographic zero-knowledge proofs so citizens can prove they were physically present within a 50-meter radius of the road damage without exposing their personal identities or daily transit routes.

Cross-Infrastructure Expansion: Generalizing the EAS schema and spatial verification engine to audit other municipal maintenance contracts, such as malfunctioning street lights (PJU), clogged drainage canals, and structural bridge maintenance nationwide.