Show HN: Resin, fast native scheduler for LLM inference engines (github.com)

Resin, a new library for building blocks in large language model (LLM) inference, was recently unveiled, featuring a high-performance Rust scheduler at its core. Unlike typical inference engines, Resin is designed for flexibility, allowing developers to create model-specific kernels optimized for their hardware. This architecture enables advanced inference capabilities, such as "vibe coding" complex models like DeepSeek-V4.1-Flash and GLM-5.3-Flash in just 12 hours, demonstrating that it can rival established engines like vLLM and SGLang. For the AI/ML community, Resin's modularity and performance optimizations offer a significant leap in customization for LLM applications. It provides features like shared kernel caching, user-defined batch processing, and powerful benchmarking tools through its HTTP server and detailed metrics support. While still in a non-production state, Resin opens the door for further innovation in model inference systems, tailored for researchers and developers who want to push the boundaries of LLM implementation across various hardware configurations.