Executive summary
The client, a global leader in smart building solutions, wanted to transform their existing AI platform for computer vision applications into a containerized, microservices-based modern platform. The aim was to enable analytics and visualizations, API-based configuration, media streaming, secure data storage, and user authentication all while interfacing with video surveillance and access control infrastructure.
eInfochips leveraged eInfochips NomAIzo™ MLOps framework for optimized model deployment, inference, analytics, and monitoring, and conducted an architecture assessment of vision AI pipelines and pre/post processing applications. The platform was rebuilt using NVIDIA’s TensorRT, Triton, and DeepStream stack alongside Kubernetes, Helm, and open-source AI frameworks.
The MLOps transformation delivered a 3x performance improvement, enabled 3x more camera streams, reduced price per channel by 70%, and supported 14 AI analytics use cases per hardware unit.
Project Highlights
- Leveraged eInfochips NomAIzo™ MLOps framework for optimized deployment, inference, analytics and monitoring.
- ML models ported to TensorRT, executed via DeepStream and Triton inference servers.
- 3x performance improvement with 14 analytics use cases per hardware unit.
- 70% reduction in price per channel through AI/ML model and pipeline rearchitecture.
- Portable coding with full Cloud-to-Edge support and scalability for additional AI/ML models.
