Developed a high-performance gRPC server for the NVIDIA/TensorRT-LLM repository, enabling seamless integration with external routers such as those implemented in Rust. Focused on backend and API development using Python and gRPC, the work introduced support for pre-tokenized input and raw token ID output, streamlining end-to-end processing within TensorRT-LLM workflows. This approach reduced routing latency and improved interoperability, allowing scalable, high-throughput inference pipelines to connect efficiently with external systems. The feature addressed the need for accelerated processing and flexible integration, demonstrating depth in backend architecture and protocol design while enhancing the overall extensibility of the TensorRT-LLM platform.