PHANTOM
Satellite-powered ghost infrastructure detection cross-referencing government claims against satellite imagery.
The problem
Billions of rupees in Indian public infrastructure funds disappear into projects that exist only on paper. PMGSY has sanctioned over ₹2.7 lakh crore for rural roads, yet in remote states like Meghalaya, some roads marked 100% complete in government databases simply do not exist. Traditional audits require inspectors to travel dangerous, inaccessible terrain and are easily falsified with staged photographs. There is no scalable way to verify if a road was actually built.
Our solution
PHANTOM cross-references government PMGSY road completion claims against Sentinel-2 satellite imagery using a U-Net semantic segmentation model. The system scrapes project data, downloads cloud-free satellite tiles, runs road detection entirely on-device using an AMD Ryzen AI NPU, and presents discrepancies on an interactive map, marking ghost roads in red, uncertain cases in amber, and verified roads in green.
What we built
- Playwright-based PMGSY portal scraper for project claims and GPS coordinates
- Sentinel-2 L2A multi-spectral satellite imagery via STAC API
- U-Net with ResNet-34 encoder for semantic road segmentation
- AMD Ryzen AI XDNA NPU for fully offline, on-device inference
- RED/AMBER/GREEN discrepancy classification engine
- Interactive Leaflet.js dashboard with satellite overlay and evidence markers
- Runs entirely offline, no cloud, no data privacy risk, battery-powered field use
The part where we talk
If you've read this far,
we should probably talk.
We partner with teams and institutions ready to build something real. Tell us what you're working on, no pitch deck required, no pressure. Just a real conversation about whether we're right for each other.
Usually a reply within a day · Delhi NCR, India