Skip to main content
Back to projects

Production CV

Production Video Analytics for Mines

Large-scale mine safety analytics across 7 deployed sites.

Production CCTV analytics across mining, energy, banking, and retail, including mine safety detection, operations monitoring, edge inference, and camera-fleet workflows.

Reference article and anonymized case study available.

Visual model

Workflow map

01Existing CCTV
02Local inference
03Cloud dashboard
04Team response

Reference visuals

Source: Mining reference article
DeepSight mining video analytics dashboard from LinkedIn article
Mining analytics dashboardDashboard view used to track safety and compliance events across deployed mine sites.
Real-time image capture popup from mining video analytics article
Real-time image evidenceDetected events are captured as timestamped image evidence for validation and response.
Real-time video snippet interface from mining video analytics article
Real-time video evidenceAlert workflows include short video snippets so teams can verify events before acting.

Problem

Coal mines needed real-time visibility into physical activity, safety compliance, restricted-area access, fire risk, traffic flow, and response workflows across existing camera infrastructure.

What I built

Led delivery of multi-site analytics including PPE compliance, fire and smoke detection, human and vehicle intrusion, vehicle counting and classification, tailgating, crowd detection, over-manning, idle-time alerts, ATM analytics, HPCL wagon tracking, and national exam workflows scaling to more than 10,000 cameras per exam.

What was hard

The difficult work sat between model accuracy and deployment reality: old IP and analog cameras, hybrid local/cloud dashboards, event routing by priority, throughput, edge constraints, camera quality, alert usability, and reliable behavior across different sites.

Result

The mining deployment raised more than 1.4 million AI events across 7 mines, supported team-based response workflows, and reported safety and compliance improvement into the 90-95% range.

What I would improve next

Continue pushing toward better monitoring, calibration workflows, root-cause dashboards, and deployment playbooks that reduce field iteration time.