Changsu Lee developed a high-performance gRPC server for the NVIDIA/TensorRT-LLM repository, focusing on seamless integration with external routers such as those written in Rust. Leveraging Python and backend development expertise, Changsu enabled the server to accept pre-tokenized input and return raw token IDs, streamlining end-to-end processing and reducing routing latency in inference pipelines. The technical approach emphasized API development and interoperability, allowing TensorRT-LLM to support scalable, high-throughput workflows. Although the work spanned a single feature over one month, it addressed a specific integration challenge and demonstrated depth in designing efficient, production-ready backend systems for machine learning infrastructure.