Week #6: Putting The Pieces Together

1. Dialogue manager

This week, the team started working on the module that sits at the core of the solution, the Dialogue Manager, responsible for receiving the user’s utterances, interacting with the machine learning models in order to extract information from them and then executing the appropriate actions, such as coming up with product recommendations for the user, which are then sent back to the user.

Besides this, the module is also responsible for maintaining the conversation’s state/context, which in turn allows the bot to adapt its personality and expressiveness based on the user’s feedback and past utterances.

This is easily achieved with GPT-3 by dynamically updating the dataset with the personality that the bot should have and by adjusting hyperparameters such as temperature, which determines the expressiveness of the model.

Despite only having implemented a small set of features, the team is satisfied with the work done on this module, as every module of the chat bot was successfully integrated and without much effort.

In the future, as the need for more features grows, the team is also confident that in its current implementation, the dialogue manager is modular and flexible enough to easily accomodate the addition of new features and modules.

2. Scientific article

This week, the team moved on to the next chapter in the scientific article, the Development chapter, where an overview of the steps taken to develop the solution is presented, all the way from the business and data understanding to the implementation and evaluation of the solution, loosely following the CRISP-DM methodology.

3. Week retrospective

This week, the team makes yet another positive balance of the work done, as all the goals were attained and the end result of the dialogue manager can only leave a sense of relief on the team, as all the modules were successfully integrated, without any noteworthy complications.

Next week is the last week of the project, before the demonstrations in week #8. Thankfully, the current state of the project leaves the team confident that everything will finished as expected, as only the user interface is left to be implemented in the solution and the conclusion to be written in the scientific article.