A new paradigm of AI is represented by LLMs (Large Language Models), which have distinct use cases and requirements compared to traditional machine learning tools. The concept of feature stores, a system meant to centralize and serve ML features to models, has been explored in the context of LLM applications. Feature stores can provide real-time pipelines of features for inference, enabling real-time personalization of LLMs. This is particularly relevant as LLMs are "reasoning engines" that require access to fresh data about users or businesses to make informed decisions. The construction of prompts in language model applications has also been discussed, with four main strategies: hard-coded prompt strings, incorporating user input, incorporating basic real-time system data, and incorporating external data fetched based on user input. Feature stores can supercharge prompt construction by allowing for the inclusion of real-time, complex, and user-based information. An example of how feature stores can be used to construct prompts is demonstrated using the Feast GitHub repository, where real-time, user-specific information is fetched and used to construct a prompt that is then passed into the language model.