Three Vs are commonly used to describe challenges in dealing with Big Data: Volume, Velocity, and Variety. There has been a lot of focus on dealing with volume (e.g. horizontally scalable NoSQL solutions) and velocity (e.g. real-time streaming solutions) aspects of big data whereas there has been no satisfactory solution for handling the variety of big data. In this talk, we will explain how semantic technologies and graph databases can be used to address the schema variety problem by providing a flexible data integration framework. Graphs can be used to represent various kinds of heterogeneous data and the RDF graph model can be used to serialize, query and reason with semantic graphs using W3C standards. Our talk will describe how we extend the popular open-source MapReduce framework Spark with RDF and SPARQL support. This extension builds on top of the Stardog RDF database we develop but provides a unified view over RDF model and property graphs as they are supported in the TinkerPop stack.