Type any drug-interaction question. NXPU answers with
UNSAFE + proof tree,
NOT_DERIVABLE, or stays silent. An LLM answers the same question fluently and
confidently — including when it doesn't know.
This page runs the chip's exact rule-firing semantics in your browser (the JS engine
reproduces what
rule_eval.v does on silicon for the 12-fact drug KB).
For byte-exact silicon validation, see the
recorded transcript of
python -m nxpu.demos.llm_vs_nxpu
running against the FPGA, or run it yourself from the
repo.