Data as codewords is the solution to the problems building real-time edge computing applications. Unlike traditional compression or encryption that operates on a file, Atombeam applies offline Al and machine learning to identify patterns in the data, and from this process we build a table of patterns and codewords that represent them. The coding process then replaces each pattern by its associated codeword, in a way that is always and completely lossless, so the actual data is never exposed but is fully and accurately represented. Even small messages that share a lot of data are "compacted" to achieve a very high level of efficiency. While the value of each codeword selection uniquely represents each source pattern, codewords are assigned randomly, not derived from the pattern. That means you cannot deduce patterns from codewords.