Worked on the kvcache-ai/sglang repository to deliver the DeepSeek V3.2 Encoding Upgrade, focusing on updating the encoding module and chat-serving logic to support a new encoding format. This upgrade enabled more robust function call parsing and smoother tool integration within chat-based workflows. The work involved backend development and API design using Python, with careful attention to data parsing and unit testing to ensure reliability and maintainability. Collaboration was emphasized through co-authored commits and clear documentation, laying the groundwork for future enhancements. The changes improved the extensibility of the system and contributed to faster, more reliable feature delivery.