A proof-of-concept that embeds a compiled llama.cpp model into a PDF, demonstrating that LLM inference can run entirely inside a PDF file.