Amazon has the machine learning service Codeguru presented as part of its web services. This should automate so-called code reviews, ie the checking of newly written code, as far as possible and give the programmers recommendations for improvement. The use of Codeguru is supported in the Github code-hosting service as well as in Codazommit's proprietary code-hosting offering.
The improvements proposed by Codeguru are, as mentioned, based on machine-learning models. To train them, Amazon uses different data sets. According to the company, this includes hundreds of thousands of internal Amazon projects and more than ten thousand open source projects. Similarly, tens of thousands of Amazon developers have helped improve the service. One of the most important categories of Codeguru guidance from Amazon's point of view is directly related to Amazon's recommended best practices for AWS SDKs and APIs.
In addition to the code review Codeguru also offers a so-called profiler. This is to find those code components that are particularly critical to the performance of an application. The service then recommends ways to reduce CPU usage, reduce the cost of computing, and ultimately improve the performance of the application. Specifically, Amazon enumerates overproduction of large objects, costly de-serialization, the use of less efficient libraries as dependencies, or simply excessive logging.
Codeguru is currently still in a preview phase and is free for testing for the first 90 days. After that, the service should be according to Amazon so inexpensive to use that it can be used for any review and also the profiler should be able to run along with the application.