Week #7: Almost There!

1. Fine-tuning

Having finished the back-end implementation last week, the team took the opportunity to implement some fine-tunings on every module this week.

On the one hand, the machine learning models were tweaked in order to offer better performance (e.g. improved the GPT-3 dataset, optimized the LSTM/GRU architectures).

On the other hand, the dialogue manager and recommendation engine were refined in order to better adapt to the client throughout the conversation, based on the feedback received during it, and to offer better and more relevant recommendations, based on the client's needs.

2. User interface

The fine-tunings referred previously ended up taking more time than expected from the team, which ultimately led to the decision of not implementing a single page application like in the previous challenges.

Instead, a simpler command line interface was implemented, which essentially serves the same purpose: allow the user to interact with the digital sales assistant and check the recommendations that are done over time by the assistant, based on what is being said during the conversation.

2. Scientific article

With only the conclusion left to be written, the team also took the opportunity to review what was written before and obtain some feedback from the teachers, with the goal of refining the article as much as possible, in order to meet the quality of the project it was written about.

3. Week retrospective

In retrospective, despite the decision not to implement a single page application like in the previous challenges, the team is happy with the final solution and scientific article, which provides the necessary confidence to tackle next week's demonstrations.