you can run the english/japanese translator aready https://git.smesh.lol/transdb it uses the cayley tree/bethe latrice as the coordinate system and the translator is about 80% accurate by sample testing, and can carry the semantic relations between ideas in such a way that it walks around in it. similar ideas are short distances. synonyms like language translation are close neighbours. it also has a basic word order and transforms, it can find bridges through other languages to find appropriate expansions.

so i am sure it cad do inference. eventually, without LLM to interpolate. because the database engine, iskradb is fully dynamic, it can learn from the LLM and user input also