Developing high-performance applications that interact with databases is a difficult task, as developers need to understand both the details of the language in which their applications are written in, and also the intricacies of the relational model. One popular solution to this problem is the use of object-relational mapping (ORM) libraries that provide transparent access to the database using the same language that the application is written in. Unfortunately, using such frameworks can easily lead to applications with poor performance because developers often end up implementing relational operations in application code, and doing so usually does not take advantage of the optimized implementations of relational operations, efficient query plans, or push down of predicates that database systems provide. In this paper we present QBS, an algorithm that automatically identifies fragments of application logic that can be pushed into SQL queries. The QBS algorithm works by automatically synthesizing invariants and postconditions for the original code fragment. The postconditions and invariants are expressed using a theory of ordered relations that allows us to reason precisely about the contents and order of the records produced even by complex code fragments that compute joins and aggregates. The theory is close in expressiveness to SQL, so the synthesized postconditions can be readily translated to SQL queries. Using 40 code fragments extracted from over 120k lines of open-source code written using the Java Hibernate ORM, we demonstrate that our approach can convert a variety of imperative constructs into relational specifications.