A code transaction modifies a set of code files and runs a set of tests to make sure that it does not break any of the existing features. But how do we choose the target set of tests? There could be 1000s of tests. Do we run all those tests? In that case, are we not wasting resources, and delaying the whole process of merging a code change? Can we choose a static set of tests? In that case, we have a chance of missing a set of tests which may be related to the code changes. How do we find out the optimal set of tests for a given code change – is the goal which we are trying to solve using the Recommendation System.