Eureka generates random inventions by analyzing the word frequencies of US patent application abstracts. By implementing a Markov processor, it produces pseudo-readable text based on US patent abstract source material. In effect, Eureka is able to babble in the language used in patent writeups.
All patent information was taken from Data.gov. Eureka was designed and built as an entry to Sunlight Labs' Apps for America 2 challenge. All source code is available on GitHub.