In this last week of the project, the focus is on making additional improvements to your conversational agent and on testing your agent. During this assignment, you are assumed to work as a whole team to test your cooking assistant, and working in subteams on the extension of the agent (depending on the type of extensions).

Extending the Cooking Assistant

This time, we do not specify what improvements you should implement. The only requirement for this week is that you should implement at least 2 more improvements for your conversational agent (which may also involve improvements to your Dialogflow agent) that you specify yourselves. Please make sure to document these improvements in your end report. For example, you could add more useful variants of patterns to improve the agent or introduce other capabilities that remove some of the existing limitations that the agent still has (which you will have discovered while testing and evaluating the agent during this project). This and other papers might also provide you with ideas.

Evaluating the Cooking Assistant

You should perform a lot of testing to verify that the happy flow and the other capabilities (e.g., clarification, repair, etc.) that you designed for your agent are working as expected. This testing should demonstrate that the happy flow can be executed and the basic mechanics of the capabilities that you implemented are working. This does not mean that your agent will perform well already in all circumstances. In a four-week project, where you started with only a minimal conversational agent, you will not have been able to develop a perfectly working cooking assistant (what does that even mean?). So there will still be plenty of work to do to further improve the agent. This can be done by improving the capabilities it already has and/or by adding new capabilities to the agent. You should identify a number of these improvements and briefly document some design ideas in your report for each of these improvements. This may also help you to come up with some improvements that you want to implement in this last week (see previous section).

Of course, as a team, you should test your own agent. But another idea is to ask the help of another team to provide you with a critical evaluation of your assistant. This may yield other outcomes because the other team will not be familiar with the exact solutions that you came up with. They therefore will be able to perform a (limited) form of black-box testing (also called functional testing) which relies on the behavior of your agent, with no (or only limited) assumptions of what happens in between. In other words, they will be able to act more as regular users of your system and can provide you with feedback from a user’s perspective. You should ask this other team to report anything they noticed while interacting with your agent. You can also ask them to complete the small questionnaire below (use a 5-points Likert-scale and, e.g., a simple Google form to collect the data).

• The overall experience while interacting with the cooking assistant is good.

• It was easy to select a recipe.

• The conversation with the agent is engaging.

• The responses of the agent are appropriate.

• The response of the agent are useful.

• The visuals helped me understand the agent.

Week-4 Deliverables

In this week, you are desired to submit your whole work, i.e.:

• Dialogflow Agent in a .zip format

• Goal Agent in a .zip format

• End Report in a .doc/.pdf format - (See Guidelines)