Design a system for fine-tuning large language models on domain-specific datasets. The system should support dataset versioning and preprocessing, distributed training job orchestration, evaluation pipelines, and model registry with deployment capabilities for iterative improvement.