Greetings!

Sprint 2 has now ended, the sprint went pretty well. Both the internal teams achieved most of the goals for the sprint.

The goal for the automation team was to add two new automation modules. One of the new automation modules that was added in the sprint is a module for setting reminders. The remainder needs a time and message as input, the time can be either relative or absolute. The implementation right now is pretty simple but work very well. The other module that was implemented is a schedule module, it functionality is to add a calendar event to the users calendar server. The implementation works okay, but it is not connected to any third party calendar server like Google calendar. There are some problems with this module because there are so many parameters needed for it and the NLP module can not parse all of them right now. Another extra feature that got implemented this sprint is a system that asks the user about an input like time if the NLP module doesn’t find any time in the original input. This has improved the user experience a great deal and we are looking to implement more of these types of features.

For the next sprint the automation team plans to improve on the already implemented modules, like making the schedule module connect to Google Calendar and to make it simpler for us we decided that we will only support Google Calendar in this module for now. A new feature we plan to implement is a confirmation message that tells the user what the AI interpreted the message to mean, so that the user can abort if it miss interpreted the input. The last goal for the next sprint is to add a system that can fetch contacts from the users Gmail account so that the user doesn’t have to write the whole email address.

The goal for the NLP team was to develop two NLP models that can understand the meaning of the input. One of the models uses Word2Sec as the main method and it doesn’t work very well yet because we have a very small amount of training data, and it uses Deep Learning which requires a lot of data for it to work well. The other model uses a technique called dependency parsing, it works very well for simple inputs that are very similar to the training data that the NLP team created this sprint.

For the next sprint the goal of the NLP team is to improve on the NLP models that we have by training them on more data that the team will collect during the sprint. Right now there is no training data for the schedule automation module, thus that will be fixed this sprint. Another goal is to find better ways to use NLP models to interpret the meaning of the input, we have started looking at having one NLP model that is specialised for each automation module. The hope is that this will make the input to the automation modules more specific so that the automation modules can support more advance features.

/The RPA Tomorrow Team