When my daughter started kindergarten, it was a new experience for our entire family. As any parent knows, sending your child to school can be a big experience for parents and children. We quickly discovered how devoted the teachers were AND how busy they were. All parents were in the same boat; they were asking questions left and right, calling the school, sending emails, showing up in person, you name it. The staff was underwater. I thought it would be interesting to try and reduce the overall volume by using a simple texting bot to answer some or most of the questions that commonly came up. The problem to solve with this project was to make a personality that would be amenable to parents and load it with all of the information it needed to know.

The technology that I chose for this was IBMs Watson. At this time, Google and Amazon were in their infancy in terms of offering an API based NLP product and there were a few others out there that required rolling a lot of services. IBM offered one of the first NLP services and made it relatively simple for developers to get started with an NLP service. Creating the personality was a lot more nuanced. It required the proper vocabulary and the ability to also chit chat.

Since the school was a bilingual education program that focus on Spanish and Chinese programs beyond the English language instruction. As such, another goal of the project was to make the bot multilingual and offer parents the ability to use their native language to communicate with the school. This required an additional layer of logic to understand the language that was being spoken.

This was a fun project because it allowed me to really dive into NLP as a technology and the implementation considerations around personality and other behavioral facets of bots. In addition, I found it interesting to understand the different cultural nuances that this entailed and how different families may be looking for the same piece of information but their native language may affect how a simple question may be posed.