The relation extraction task is crucial in extracting structured knowledge from unstructured text. Traditional supervised approaches for relation extraction heavily relies on large labeled datasets, which are often expensive and time-consuming to acquire. Few-shot relation extraction aims to address this challenge by enabling models to learn from limited labeled data. In this research project, I propose a novel approach to improve few-shot relation extraction by leveraging knowledge graph facts for prompt creation in language models. Specifically, I exploit the rich semantic information captured in knowledge graphs to generate informative prompts that guide the model toward accurate relation extraction. Our methodology involves constructing prompts by incorporating relevant entities, attributes, and relations from the knowledge graph. Encoding this domain-specific knowledge within the prompts enables the language model to effectively generalize and make accurate predictions, even when faced with limited labeled data.