CityShield · VisionScan
Fully offline on CPU, it closes the loop from live CCTV anomaly to dispatched patrol, with physical and cyber crime on one GIS map. Built with the Cyber Crime Branch of Ahmedabad City Police for KANAD S.H.I.E.L.D. 2026.

Highlights
- Four CCTV search modes built on CLIP with FAISS, ArcFace and YOLOv8.
- Object tracking with zone and line-crossing analytics.
- Generates 7 statutory documents in English, Hindi and Gujarati, with a per-user language interface.
- Ships a 4-layer OWASP middleware and 77 passing backend tests.
- Risk model backtested on synthetic data as a methodology demo: Hit-Rate@10 0.771, PAI@10 2.31x over 8 weekly folds.
The full story
A unified AI policing platform that fuses physical and cyber crime onto one GIS map, forecasts next-week risk, and closes the loop from a live CCTV anomaly to a dispatched patrol, fully offline on CPU. Object tracking with zone and line-crossing analytics, and a per-user English, Hindi and Gujarati interface. 77 backend tests. Built for KANAD S.H.I.E.L.D. 2026.
Four CCTV search modes (CLIP to FAISS, ArcFace, YOLOv8), 7 generated statutory documents (en/hi/gu), a 4-layer OWASP middleware, and 77 passing backend tests. Risk model backtested on synthetic data (Hit-Rate@10 0.771, PAI@10 2.31x over 8 weekly folds) as a methodology demonstration. Built for KANAD S.H.I.E.L.D. 2026 with the Cyber Crime Branch, Ahmedabad City Police.