Can Small Teams Do MLOps Too? Starting Simple Without a Big Budget

Abstract

MLOps is viewed as a complex process for the enterprise level, so MLOps can serve as a significant hindrance for small teams who want to apply machine learning operations. Yet, small teams can gain a tremendous advantage from MLOps by applying simplified, lean tooling, gradually moving toward more complex MLOps for their teams. In this article, a complete set is presented for small teams on how MLOps can be applied effectively for small teams without engaging cloud orchestration platforms. The article discusses ideas on how MLOps can be applied for small teams through basic version control for source and model files, tooling for simple experiments with file storage and databases, basic automation through shell scripts, basic MLOps tooling through system job schedulers, and basic MLOPs testing through standard testing results. MLOps can be made a successful process for small teams with the use of a progressive approach. According to the progressive approach, teams can move toward more complex MLOps concepts when their skills and resource availability increase. Therefore, in the progressive approach, automation investments can provide a remarkable difference for teams, meaning investment in MLOps can be avoided because automation can provide a negative effect for the team. Therefore, even the leanest teams can attain a solid basis for successful MLOps.