AI Nirikshak: Nagpur's pilot for AI-led crowd management goes live for Maharashtra Winter Assembly
Nagpur police have rolled out AI Nirikshak for the Maharashtra Winter Assembly Session 2025. Officials say this is India's first AI-based application in policing for crowd monitoring and management, set against recent stampede risks and security incidents like the Delhi blast.
The pilot is a joint effort between Nagpur police, a city-based private tech firm, and Microsoft. Built on Microsoft Azure's cloud, cognitive services, and video analytics, it fuses feeds from fixed CCTVs, mobile surveillance vans, drones, and external road monitoring points.
A city-based firm's orchestration engine runs TensorFlow and PyTorch models at the edge for speed and scale. The goal is simple: prevent overcrowding, enable faster decisions, protect VIP routes, and use manpower where it matters most.
What managers should notice
- Real-time crowd density with heatmap hotspots to spot rising congestion early.
- Predictive surge alerts using movement trends and multi-source data fusion to anticipate build-ups.
- Abnormal activity detection: commotion, unattended objects, sharp weapons, and restricted zone violations (e.g., unauthorised vehicles on VIP routes).
- Facial recognition against watchlists for suspects, footfall analytics at gates, and alerts for behaviours like loitering or barrier bypass attempts.
- Police AI Agent ChatBot for instant semantic search across incidents, FIRs, suspects, vehicle histories, and alerts-supporting 24x7 field and command teams.
How the system is structured
- Azure cloud with cognitive services and video analytics, integrated with on-ground camera networks and mobile units. Learn about Azure AI services.
- Edge inference using TensorFlow and PyTorch to keep latency low and performance consistent during peak loads.
- GDPR-friendly architecture, audit trails, and scalability for city-wide deployment.
Leadership and execution
Commissioner of Police Ravinder Singal led the initiative with a core team including DCPs Lohit Matani, Rushikesh Reddy, and Deepak Aggrawal, alongside AI specialists. Their vision: make Maharashtra a national leader in AI-led crowd management, beginning in Nagpur and scaling across the state.
The pilot aims to reduce crowd surge risk, enable early interventions, secure VIP routes, and streamline deployment based on real-time demand. During large sessions that attract thousands, this means better situational awareness and faster response.
Operational benefits for high-density events
- Early warnings for choke points with clear, low-latency alerts.
- Dynamic staffing: shift teams based on live footfall and predicted inflow/outflow.
- Common operating picture across control rooms and field officers with the AI ChatBot.
- Clear handling of restricted areas and VIP corridors.
Implementation playbook for managers
- Objectives and risks: define surge thresholds, restricted zones, VIP corridors, and response timelines.
- Data governance: establish watchlist approval, retention limits, and audit logs; document a DPIA and signage policy. See EU guidance on data protection principles: EU data protection rules.
- Vendors and SLAs: pair a local integrator with a hyperscaler; lock in uptime, latency, support, and edge failover requirements.
- Training and drills: simulate surge scenarios, run red-team tests, and train officers on alert triage and escalations.
- Public communication: inform citizens of surveillance zones, purpose, and redress channels.
- Incident workflows: standardise escalation matrices, cross-agency coordination, and post-incident reviews.
KPIs to track from day one
- Time-to-alert and time-to-dispatch.
- Peak density reduction at hotspots and average gate throughput.
- False positive/negative rates for key detections; watchlist match precision/recall.
- System uptime, edge-cloud failover success, and alert latency.
- Man-hours reallocated to priority zones and citizen complaint volumes.
Risks and safeguards
- Facial recognition bias and error: keep a human-in-the-loop and restrict use to defined cases with approvals.
- Privacy and security: enforce data minimisation, strict retention, encryption, and regular audits.
- Operational resilience: plan for network outages with edge processing, redundant links, and manual overrides.
- Public trust: maintain transparency, publish governance summaries, and enable independent evaluation.
What's next
If the pilot meets its targets, officials expect a statewide rollout across high-footfall events-festivals, political gatherings, and major public programs. The Nagpur deployment offers a practical blueprint other cities can adapt without starting from scratch.
For leaders building similar capabilities, upskilling your teams on AI operations and governance pays off quickly. Explore role-based training here: AI courses by job.
Your membership also unlocks:
