Jemish Mavani

• Spearheading integration of Vision-Language Models (VLMs) and LLMs into video analytics, enabling automated store audits, anomaly detection, and compliance reporting.

• Designing multi-location retail and hospitality solutions that reduce manual audit effort by 60%, transforming raw CCTV feeds into actionable insights.

• Partnering with cross-functional teams to standardize hybrid cloud + edge deployments, ensuring scalability and operational adoption.

• Safe City Deployment: Led the development of 45 CV analytics use cases (safety + retail) deployed on 300+ surveillance cameras, achieving 33% lower CPU utilization. Partnered with government agencies to ensure compliance with public safety standards and system reliability.

• Metro Safety: Delivered real-time passenger safety monitoring across 30+ cameras, including crowd management, intrusion detection, and PPE compliance, reducing incident risk and cutting operator response times by 40%. Collaborated with metro officials to refine models based on field feedback and

ensure adoption by staff.

• Warehouse Automation: Deployed object detection and activity recognition models for forklift safety, pallet dwell monitoring, and truck entry/exit analytics on Nvidia Xavier NX, improving efficiency by 25% while reducing workplace incidents. Drove adoption by aligning outputs with operational protocols and safety compliance requirements.

• Mentored junior engineers and coordinated cross-team efforts to accelerate delivery timelines and knowledge transfer.

• Developed a Face Mask Detection system with 95% accuracy, integrated into workplace CCTV, ensuring regulatory compliance during pandemic operations. Delivered a production-ready solution that was adopted across multiple offices.

• Built an Exam Anomaly Detection System using real-time CV models (face recognition, object detection) to identify unauthorized devices, multiple-user presence, or absence. Generated automated video evidence reports, significantly reducing manual proctoring effort.

• Delivered multiple client-facing proofs of concept (POCs) in object detection and activity recognition, successfully converting several into full production deployments