How to Enrich LLM Context to Significantly Enhance Capabilities

Large Language Model performance can be greatly improved by enriching their context with additional metadata. This data, such as filenames, timestamps, or folder paths, is often readily available but overlooked. Techniques for gathering this information include pre-processing documents to extract key details or using on-demand LLM calls to fetch specific data points as needed. One powerful use case is metadata filtering in search, which helps Retrieval-Augmented Generation (RAG) systems select more relevant information to answer user queries.