Coding, creatives, and conversation.

Week 3

In our third week, the class mainly focused on our A2, in which we had to actually code the conversational aspects of our conversational agent. Ours being Srishti Local Guide, primarily a navigational agent, a large part of the conversation we coded was giving directions to the user.

Initially, one of our ideas was to have image-based navigation, wherein on being asked where they were, the user would receive a series of images of landmarks, leading them to their destination — this too, was an idea that was adapted from our initial idea of sending navigational videos. We wanted the navigation to be visual as it’s easier to understand and there’s less searching to be done on the user’s behalf, but eventually we ended up going with text-based navigation.

To begin, we plotted a series of landmarks and routes that could be used as a guide, and then wrote different possibilities the conversation could explore. When coding, this involved a lot of if and else and while loops which made the conversation quite open ended, as the user could have a variety of answers. Aside from this, we had received feedback on making sure that our Guide doesn’t completely eliminate human interaction — so in certain conversations, like when the user was facing a difficulty in locating landmarks or when they were in a space full of people, the Guide would prompt them to ask someone for assistance. This was one feature I liked as sometimes we need reminders to ask others for help and not everything can be done by a machine.

Since we couldn’t link it to the student portal, we made individual folders for every user, with their class information, which the code would pick up information from when the user asked where their class was, about their faculty, etc. This personalized experience was another really enjoyable aspect, as to the user it’s practically like the Guide knows their student portal, where in actuality it doesn’t.

Going forward to our A3, I think I’d like to see how to make even the navigation aspect conversational. I think that will be challenging as generally when someone is trying to get a job done, especially if they have a time limitation, they tend to only care for the answers and not for other inputs the Guide may give.