Developed a federated multimodal learning framework within the NVIDIA/NVFlare repository, enabling privacy-preserving collaborative fine-tuning of unified models using BLIP and JanusPro backends. The work involved setting up the environment, implementing model registration, and creating end-to-end training scripts to support enterprise-scale deployments without sharing private data. Leveraging expertise in Python, data science, and federated learning, the developer ensured local quick tests passed and maintained comprehensive in-line documentation. This contribution established a foundation for secure, distributed machine learning workflows, allowing multiple clients to participate in model improvement while maintaining data confidentiality and supporting future extensibility within the NVFlare ecosystem.