A language experiment with a real toolchain behind it

On March 14, 2026, Show HN: Han reached 199 points and 106 comments. The author says the project began after seeing an AI-assisted rewrite of a C++ codebase into Rust and wondering what a programming language would look like if every keyword were written in Hangul. The result is Han, a statically typed language implemented in Rust that is framed as a side project and design exercise rather than a replacement for Python or mainstream systems languages.

The HN description and the project README outline a surprisingly complete stack. Han includes a lexer, parser, AST, tree-walking interpreter, and LLVM IR code generation path. The command-line tool exposes interpreter mode for instant execution, native compilation through LLVM IR and clang, a REPL via hgl repl, and an LSP server through hgl lsp. The README also lists arrays, structs with implementation blocks, closures, pattern matching, try/catch-style error handling, file I/O, module imports, and generics syntax as supported language features.

Why the project stands out

What makes Han more than a visual gimmick is that the Korean keywords are not just decorative aliases. Core syntax such as 함수, 만약, 반복, and 변수 is built directly into the language, and the project supports Hangul identifiers as first-class names. The README positions that as both a cultural and educational experiment: programming does not have to look English by default, and learners interested in Korean can read and write real code using native terms instead of merely localizing comments or strings.

The repository also makes the engineering choices legible. Han uses text-based LLVM IR generation instead of the LLVM C API, which keeps the build path relatively simple while still targeting native binaries. It ships with a VS Code extension, a basic LSP server, and a published test count of 46 tests across lexer, parser, AST, interpreter, and code generation. That gives the project a more serious compiler-workbench feel than most novelty-language demos.

An interesting AI-adjacent wrinkle

One especially relevant detail is the README’s tokenizer note. The author reports that Hangul-heavy Han examples use more GPT-4o tokens than equivalent Python or JavaScript snippets, and frames that as a tokenizer training bias rather than an inherent property of Korean. That makes Han relevant not only as a language experiment, but also as a reminder that current LLM tooling still privileges English-shaped syntax in subtle ways.

Original source: Han on GitHub. Community discussion: Hacker News.

#hangul

Show HN: Han Turns Hangul into a Statically Typed Rust-Built Programming Language