Software Engineering Institute

The TEC tool allows users to detect mismatched expectations among the teams involved in building an ML component. TEC helps the teams that develop, deploy, and sustain ML components to avoid mismatches that lead to unnecessary rework, project delay, poor component performance, incompatible interfaces between components, insufficient computing resources, and the inability of the ML-enabled system to detect troubling issues.

ML system development typically involves three teams with their own workflows and perspectives: data scientists or ML engineers build the ML model, software engineers incorporate the model into the software system, and operations staff move the model into production and monitor its operation. When these teams fail to communicate with one another about the ML aspects of the project, ML mismatch results due to incorrect assumptions.

With TEC, all three teams use the same set of descriptors for their ML component expectations, such as required computing resources or runtime metrics and the context of the ML-enabled system that incorporates the components (e.g., system goals and how model outputs will be used by the system). The tool compares information across descriptors and flags any mismatches or missing information. The stakeholders can then resolve problematic differences early in the development life cycle, meaning less rework and ML-enabled systems that meet mission and business goals.