RoadReport

RoadPulse is an AI-powered road inspection platform that turns any phone photo or dashcam clip into a graded, tracked, and actionable road defect report — no engineering team required. A citizen spots a pothole, snaps a photo, and submits it with their GPS coordinates. On the backend, Meta's SAM 3.1 segments and localizes the damage in the image, while Muse Spark 1.3 produces a structured assessment: damage type, severity score (1–5), and a repair priority ranking. Dashcam surveys run through Muse Wonder 1.0 for multi-frame analysis across longer road stretches. All reports aggregate into a live PostGIS-backed map where hotspots form automatically as multiple reports cluster around the same location. Road agencies get a ranked fix-first list, one-click evidence PDFs, and a natural-language analyst chat (also powered by Muse Spark) that answers questions like "which LGA has the most critical potholes this week?" — grounded in real data, never hallucinated, via a whitelisted tool-calling agent. The system is designed for trust and scale: EXIF data is stripped after read, uploads are rate-limited, confirmations are one-per-device, and all model output is Zod-validated before it touches the database. It runs fully on mock AI by default, switching to live Meta models with a single environment variable. Built for the Meta Global AI Developer Hackathon using Meta AI models, Fastify, React, and PostGIS.