Developers often collect and store large amounts of JSON data in RDBMS (e.g., when consuming logs or data from public web-APIs). However, analysts and other data practitioners lack high-performance tools for querying and analyzing this data (which doesn’t have a fixed schema). Dominik Durner et al. propose automatically detecting the implicit common structure across a collection of JSON objects to accelerate the analytical processing of JSON data. Their collection of algorithms and techniques, JSON tiles, automatically detects the most important keys, which it materializes as relational columns; and stores infrequent keys and outlier objects in an optimized binary format for fast access. They implement JSON tiles in Umbra, a RDBMS that supports in-memory analytics for flash-based storage.