works as a transparent proxy. we extract lexical/syntactic features and compute personality dimensions without calling another LLM. fully reproducible scores, no judge variability.

use case: you want your support bot to stay "warm and professional." lindr tells you when it's drifting toward "cold and robotic."

happy to go deep on the methodology if anyone is curious.