Worked on the opensearch-project/ml-commons repository to deliver Agentic Memory Structured Output for fact extraction, focusing on enabling strict JSON schema enforcement and reliable downstream parsing. Developed constrained decoding techniques to ensure consistent and valid JSON responses, particularly for Bedrock Converse integration. Enhanced backend robustness by introducing an isJsonObject guard, refactoring parameter injection into helper methods, and establishing a constant for JSON enforcement messaging. Updated connectors and added blueprints for Coherence v2 and Gemini, ensuring schema consistency across all responses. Utilized Java for backend development, emphasizing API development, JSON handling, and comprehensive unit testing to maintain code quality and reliability.