The growing popularity of the CSV and JSON format has fueled increased interest in loading and processing CSV/JSON data within analytical data
processing systems. However, in many applications, parsing of this data dominates performance and cost. In this lecture, we introduce Mison, a novel CSV/JSON parser designed to enhance the efficiency of loading and processing data in analytical data processing systems. Unlike traditional parsers built using finite state machines, Mison employs a two-level approach. At the upper level, it predicts the logical locations of queried fields based on patterns in the dataset, while at the lower level, it constructs structural indices on CSV/JSON data to map logical locations to physical ones. This approach eliminates the need for costly tokenizing steps and transforms control flow into data flow which allows for enhancing performance by leveraging parallelism of SIMD units in modern processors. Experimental evaluations with real-world datasets and the TPC-H benchmark demonstrate significant performance benefits over existing parsers, with some cases showing improvements exceeding one order of magnitude.