Knowledge graphs have proven to be a highly useful technology for connecting data of various kinds into complex, logic-based models that are easily understood by both humans and machines. Their descriptive power rests in their ability to logically describe data as sets of connected assertions (triples) at the metadata level. However, knowledge graphs have suffered from problems of scale when used against large data sets with lots of instance data. This has by-and-large hampered their adoption at enterprise scale. In the meantime, big data systems (using statistics) have matured which can handle instance data at massive scale – but these systems often lack in expressive power. They rely on indexing which is often incomplete for solving advanced analytical problems. LeapAnalysis is a new product that married these 2 worlds together by utilizing graph technologies for metadata, but leaves all instance data in its native source. This allows the knowledge graph to stay small in size and computationally tractable, even at high scale in environments with billions of pieces of instance-level data. LeapAnalysis utilizes API connectors that can translate graph-based queries (from the knowledge graph) into other data formats (e.g., CSV, Relational, Tabular, etc.) to fetch the corresponding instance data from source systems without the expensive step of migrating or transforming the data into the graph. Data stays truly federated and the knowledge graph is virtualized across those sources. Machine Learning algorithms read the schema of the data source and allow users to quickly align those schemas to their reference model (in the knowledge graph). Using this technique, graph-based SPARQL queries can be run against a wide range of data sources natively and produce extremely fast query response times with instance data coming in as fragments from multiple sources all in one go.