Engineering AI systems
that operate at scale.

The platform operates across 8,000+ ATM sites, creating a distributed environment where video, alarms, telemetry, and operational events have to move reliably between remote infrastructure and centralized services. My work involved building and integrating backend components capable of handling this distributed workload across cloud and containerized environments.

Security events need to move from remote systems into processing and operator-facing services with minimal delay. I worked on the event-processing and communication layers using Apache Kafka, WebSockets, and REST APIs, connecting incoming security events with downstream services and real-time interfaces.

My primary focus included the computer-vision layer. I trained and deployed YOLO and PyTorch-based vision models for security-event detection and worked on the inference pipeline that processes live RTSP video and converts visual observations into structured events. The challenge was making inference reliable enough to operate continuously within a production video pipeline.

Production surveillance also requires integrating physical security hardware with software systems. I worked on hardware research, device integration, and the communication layer connecting legacy intrusion panels with backend services, including implementation around the SIA protocol.

The system is composed of containerized services deployed across multiple environments. I worked with Docker, GitHub, and CI/CD workflows to automate deployment across production infrastructure (AWS, Azure, and OCI). Beyond the cloud, I also manage the physical networking and remote administration of GPU-heavy industrial servers required for intensive edge inference.