Neurastheniacc: LLM-based C->WASM compiler you didn't need (github.com)

Neurastheniacc has introduced an innovative C to WebAssembly (WASM) compiler that leverages the capabilities of large language models (LLMs) to perform the C to WASM translation process. Unlike traditional compilers, neurastheniacc employs an LLM to convert a restricted subset of C code into WebAssembly Text (WAT) format, which is then compiled into bytecode. This two-pass approach is significant as it eliminates the need for extensive compiler expertise, ultimately making the programming experience more accessible to developers. The compiler supports a compact set of features, allowing for functionality such as nested blocks and uint32_t returns, but notably excludes pointers, arrays, and other complex data structures. The neurastheniacc compiler runs completely locally, requiring no API keys or external inference calls—only the necessary Qwen model weights. This independence not only enhances user privacy but also allows developers to fine-tune the model or adjust various parameters to optimize compilation. Initial tests have shown promising results, with high pass rates when the appropriate model is utilized. The flexibility in model selection and configuration, combined with the potential for improved performance, positions neurastheniacc as a disruptive tool in the AI and ML landscape, particularly for developers seeking efficient ways to compile C to WASM without traditional barriers.