Operationalization of Data Science (MLOps)

Operationalization of AI processes

Using Artificial Intelligence to gain insights from data is already a great advantage. This advantage can be multiplied a hundredfold if the AI process is embedded in the actual operational process and can thus constantly deliver its added value. This in turn requires the AI process to operationalize itself: AI must be available as a service, via REST API, as a batch job.

Provisioning and monitoring of services

The provision of AI services in the required form must be agile, performant and traceable. The same requirements apply as for all other business-critical processes.

Long-term technical and business monitoring (BIAS) of AI models.

AI processes are subject to a constant learning and aging process, as the underlying business process is also evolving. A progression in the operational process must be countered with an evolution of the AI process and this must be automated as much as possible. Likewise, potentially undesirable influences ("BIAS") must be detected and documented.

Data cataloging/governance

Central cataloging of data enables faster and more agile development of AI processes. When the origin and meaning of data is documented, AI can make the right predictions more safely and reliably.

Data Virtualization

Making data available to the AI process can be a challenge. Data virtualization can help by making data available to the AI process not physically, but virtually, fully documented and protected.