Techniques are provided for generating specifications for a microservice implementation of an existing application. An exemplary method comprises: analyzing request data and corresponding response data for an application implemented as a monolithic application and/or a Service Oriented Architecture application to generate data features; parsing an audit log and/or a transaction log of the application to identify interactions with a data store; clustering the data store interactions using an unsupervised learning technique to identify patterns of usage of the data store; selecting one or more service types to generate using a trained supervised machine learning model for the requests, the corresponding response data and the data store interactions; and generating an application programming interface specification, a data model specification and/or a message specification for the selected service types for a microservice implementation of the application. A run-time environment, a data definition language and/or message queues are optionally generated for the one or more selected service types.