Worked on reliability improvements for Vertex AI embeddings within the langchain4j repository, focusing on authentication consistency across deployment environments. Addressed a bug by implementing unified GoogleCredentials propagation, ensuring both PredictionServiceSettings and LlmUtilityServiceSettings use the same credentials provider. This approach removed the dependency on Application Default Credentials, preventing token-counting failures in environments lacking ADC. The solution was validated through comprehensive manual and integration testing, resulting in a non-breaking change with no API modifications. Collaborated closely with another contributor, leveraging Java and backend development skills, as well as API integration expertise, to enhance embedding reliability and deployment consistency for users.