Bayshore developed and deployed two models that are used to solve these two separate problems. Both are transformer-based models which have been shown as quite adept at language tasks. One model has been fine-tuned to extract key sentences from a given input text and the other to change the sentence structure (by means of using similar words, changing the syntax, etc.) while retaining the meaning. In both scenarios, the input texts are pre-processed in the backend before feeding into the model. Similarly, after the model outputs the desired text, post-processing steps are implemented to ensure sentence completion and reduce grammatical errors.
To include the feature of uploading a document (Pdf, doc, txt) instead of manually inserting the input text, a simple OCR (Optical Character Recognition) module has been integrated as well.