A developer trained a logistic regression and SVM classifier using labels generated by an LLM. This hybrid approach uses a large model to bootstrap a smaller, faster classifier for comment moderation. It demonstrates a practical pipeline for distilling LLM intelligence into lightweight statistics. Practitioners can use this to reduce inference costs for simple classification tasks.