Worked on reliability and model loading improvements across two repositories, netbirdio/netbird and kvcache-ai/ktransformers. In netbirdio/netbird, addressed TLS handshake failures by enabling HTTP/2 support in the Caddyfile, restoring secure HTTPS-only connections on port 443 and reducing connection errors for enterprise users. For kvcache-ai/ktransformers, enhanced Qwen3.5 model loading by refining VL model handling, updating key paths, and consolidating SafeTensor loader code, which improved maintainability and stability for FP8 model experiments. Leveraged Python, Shell scripting, and configuration management skills to deliver targeted fixes and features, focusing on operational reliability and code quality in networking and machine learning workflows.