We do natural language processing with some custom tokenisation, the data it uses at runtime is put into an sqlite db at build-time. Works fine for our usecase.