As software applications grow increasingly complex, particularly in their input formats, testing these applications becomes a challenging endeavor. Automated testing techniques, such as search-based white-box fuzzing, have shown promise in addressing these challenges. However, generating well-formed inputs for fuzzing remains a significant obstacle. In this paper, we present novel techniques as an academic proof-of-concept for automatically inferring JSON-based schemas to enhance search-based white-box fuzzing, focusing on Java and Kotlin applications. Our work offers an alternative approach to black-box grammar-based fuzzing.