What I Built
Municipal road maintenance in urban India faces a fundamental economic bottleneck: traditional road quality assessment relies either on Network Survey Vehicles (NSVs) equipped with expensive laser profilometers (costing upwards of ₹50–75 Lakhs per vehicle) or on reactive citizen complaints filed weeks after a pothole has already caused accidents or vehicle damage.
FLINT (Urban Sensing Platform) was built to invert this paradigm: Zero Fleet Cost Infrastructure Monitoring.
Instead of deploying specialized survey fleets, FLINT transforms everyday public transit buses into continuous, autonomous road-sensing units. As municipal buses travel their standard passenger routes, standard dashcams running lightweight edge AI scan the asphalt in real time, detecting pavement defects, measuring road roughness, and instantly generating GIS-tagged municipal work orders with automated Bill of Quantities (BOQ) repair cost estimates.
Key Capabilities
Dual-Head Onboard Edge Pipeline:
Real-time road distress detection (potholes, alligator cracks, waterlogging) and dynamic traffic density tracking.
Ground ROI polygon masking and Optical Flow surface verification to filter out camera shake, shadows, and overhanging tree foliage.
ByteTrack multi-frame association to eliminate duplicate detections of the same pothole across sequential frames.
Automated Municipal Triage & BOQ Costing:
Translates detected surface defects directly into municipal engineering deliverables.
Calculates patch volume, asphalt tonnage (cold mix vs. hot mix vs. full-depth reclamation), and labor overhead using standard Public Works Department (PWD) schedule of rates.
Generates automated municipal work tickets (e.g., PWD-RAI-2026-1042).
Multimodal VLM Visual Audit (Google Gemini):
High-severity defect frames are audited by Google Gemini 1.5 Flash, evaluating structural hazard severity, edge spalling, and moisture penetration before dispatches are committed.
GIS Command Center & Synchronized Video Vault:
High-framerate MapLibre GL WebGL map displaying live bus movement (BUS-01), dynamic Google Traffic-style road condition heatmaps and route reports.
Video Vault synchronizing 1080p bus dashcam playback with GPS telemetry, allowing municipal engineers to scrub directly to the exact millisecond of any road defect.
Demo
Live Application: flint-dev.vercel.app
Partner Technologies
- Google Cloud / Google Gemini API (Multimodal VLM)
To prevent municipal work crews from being dispatched on false positives (e.g., road paint, tire skids, manhole shadows), FLINT integrates Google Gemini 1.5 Flash as an automated secondary auditor:
High-severity pavement distress crops are packaged with localized road metadata (coordinates, timestamp, estimated area) and sent to Gemini via structured prompt engineering.
Gemini assesses structural severity, detects moisture pooling, and returns a verified classification, confidence score, and remediation suggestion:
Hackathon Experience
Building FLINT was an adrenaline-filled, end-to-end civic tech sprint. What made this hackathon memorable wasn't just writing code, but solving real-world edge cases that only emerge when physical hardware and geospatial data collide
Top comments (0)