Worked on the spring-projects/spring-ai repository to address a critical issue in OpenAI-compatible streaming responses. Focused on backend development using Java, the work involved implementing a guard to ensure only non-empty tool_calls were processed during real-time streaming. This technical approach prevented the misclassification of text chunks and preserved all text fragments, resolving a bug that previously caused streams to collapse to the first fragment. The solution improved reliability for users by reducing visible errors and maintaining the integrity of streamed content. Collaboration with another contributor ensured the fix was robust and aligned with the project’s standards for Java-based backend systems.