WASM-based instant code transforms for simple tasks, achieving 352x speedup over LLM inference with zero cost.
