Teaching a Web Page to Read a Chess Board A developer built a browser-based chess board reader that identifies a position from a screenshot using a 131KB file of signed bytes and roughly fifty lines of JavaScript, running inference in about a quarter of a second with no machine learning runtime. The project, part of the chessreport.ca board editor, went through four neural networks over about three days, and the developer says most of the work was not the neural network. The image never leaves the client, avoiding server inference load and keeping users' screenshots private. This entry has three interactive demos that need JavaScript, and they run the real