Distillation by Prompt: On-Premise LLM Calibration

An MSU/UA PATENT Lab paper describing the principles of ROC driven prompt distillation. It closes the gap between frontier models and on-premise LLMs allowing for offline deployments, without any cloud vendor dependencies. In my example it is used to train an installment loan lead underwriting filter.

Continue reading “Distillation by Prompt: On-Premise LLM Calibration”

LLMZ+: Contextual Prompt Whitelist

MSU research paper to be presented at a ICMLA ’25 conference in Boca Raton, FL. It describes the principles of contextual whitelisting for agentic LLMs. It significantly increases deployment security without additional cost and resources associated with traditional threat mitigation systems.

Continue reading “LLMZ+: Contextual Prompt Whitelist”