This study explores the use of large language models (LLMs) to reduce medication direction errors in online pharmacies. It introduces MEDIC (medication direction copilot), a system that integrates domain knowledge and LLMs to prioritize precise communication of core clinical components of prescriptions, such as dosage and frequency. MEDIC fine-tunes a first-generation LLM using 1,000 expert-annotated and augmented directions from Amazon Pharmacy to extract and assemble complete directions with safety guardrails. Compared to two LLM-based benchmarks, MEDIC recorded significantly fewer near-miss errors, reducing them by 33% in a production environment. The study highlights the importance of combining LLMs with domain expertise and safeguards to improve the accuracy and efficiency of pharmacy operations, enhancing patient safety.This study explores the use of large language models (LLMs) to reduce medication direction errors in online pharmacies. It introduces MEDIC (medication direction copilot), a system that integrates domain knowledge and LLMs to prioritize precise communication of core clinical components of prescriptions, such as dosage and frequency. MEDIC fine-tunes a first-generation LLM using 1,000 expert-annotated and augmented directions from Amazon Pharmacy to extract and assemble complete directions with safety guardrails. Compared to two LLM-based benchmarks, MEDIC recorded significantly fewer near-miss errors, reducing them by 33% in a production environment. The study highlights the importance of combining LLMs with domain expertise and safeguards to improve the accuracy and efficiency of pharmacy operations, enhancing patient safety.