Developed LoRA support with speculative decoding for vLLM inference in the IBM/vllm repository, focusing on enhancing efficiency and scalability for LoRA-based deployments. Integrated Low-Rank Adaptation into the speculative decoding workflow by updating inference classes and implementing a batch inference validation script to ensure robust batch processing. Addressed compatibility issues through a targeted bugfix, aligning with project requirements and maintaining comprehensive test coverage. Leveraged deep learning and machine learning expertise, utilizing Python for both implementation and automated testing. The work enabled more efficient handling of LoRA requests in batch workflows, supporting scalable and reliable inference in production environments.