Whooo’s Reading, New York, NY
Whooo’s Reading (WR) is an education technology (edtech) start-up founded by a group of Williams alumni from the Class of 2012 with the goal of creating and promoting better reading comprehension assessment tools for teachers while maintaining a fun virtual learning environment for students. Their main product is a site where students can log on to take open response and multiple choice reading comprehension quizzes, change their owl avatars, and redeem points earned for rewards in the Owl Shop and teachers can log on to monitor students’ progress and engagement, class quiz scores, and modify student assignments. Sticking with the owl theme of the company name, the user is guided through the site and demos by an avatar named Owlfonso. This summer at Whooo’s Reading was a fantastic experience to learn about myself as a programmer, add technical skills to my repertoire, and see what it looks like to be part of a team of developers.
Over the course of the summer, I completed three projects for the dev team: I used a data set of student answers to train an existing natural language processing (NLP) model to recognize book title usage; helped automate the process of creating fill-in-the-blank questions; and built a Slack app that allows Whooo’s Reading team members to pair with students to gain real-time anecdotal product usage. I worked on the first of the three projects as my main project over the course of the summer, a project motivated by the need to help the site’s automatic grading system recognize when a student is referring to a book by its title—which may confuse the model if the words in the title would not be grammatically correct in the context of the sentence. I started with learning about an out-of-the-box NLP model that could be used to parse English sentences into clauses, words, and sentence trees. By looking through examples of student responses and the dependency trees generated by this model, I determined a (non-exhaustive) set of rules for when students refer to a book by its title. With these rules in place, a large dataset of tens of thousands of student responses could be filtered to include the highest possible percent of book references. This dataset, in hopes of collecting a set of student responses annotated with the location of the book title to later train the NLP model on, then could be fed through a crowd-sourced data collection platform in a task I designed. Workers on that site could earn money from WR for each passage they highlighted book references in. From each submission, I extracted the location of each highlight, which served as the information needed to train the model to recognize book titles. Essentially, show the model as many examples as possible of a proper uses of a book title in a sentence (as well as sentences that did not include a title reference), and it could learn to detect them on its own. Because I was using an existing NLP model, I found that the most important piece of this training was having a clean dataset.
While the first of the three projects spanned the largest arc of my eight-week internship, the other two projects provided additional challenging projects to round out my experience. Because the site offers quizzes on tens of thousands of books that students of all reading levels across the country are reading, it would be inefficient for the WR team of ten to write fill-in-the-blank questions for each book. Instead, we filtered answers that students have submitted to open response questions to extract a set of pertinent, factual sentences about each book and then removed the key words from each sentence, hoping to create at least three questions for each book. This project was launched in response to feedback from teachers who would like to have easier questions that assess whether or not a student has read the book in addition to assessing their analytical skills. The final project I worked on, in my last week with Whooo’s Reading, was a fun extension that required connecting Slack (the communication app used by the WR team) to incoming quiz results from the WR site. The bot I created kept track of a few students for each team member, sending a DM through to the team member each time one of “their” students submitted a quiz or purchased something in the Owl Shop on the site. The user could also prompt the Slack bot with a student’s ID (a randomized pair of words to maintain student anonymity) and receive a recap of the student’s activity.
Overall, I was very pleased with the way my projects and workload turned out. While some of them were very independent in nature, I communicated regularly with my mentor, asking questions and getting feedback and other guidance when stuck. I also virtually connected regularly with other members of the organization, hearing about their projects and how they all fit together. I very much enjoyed connecting with a couple of the company members individually to hear their favorite parts about working in edtech and how they have grown with the company. I learned a lot about start-up culture, collaboration, and technical components of creating a machine-learning based product.
Coming into this internship with little to no expertise in the areas of Natural Language Processing, Machine Learning, and crowd-sourced data collection, I was introduced to a subfield of computer science that I remain very interested in, which will help direct my focus over the next two years in the computer sciences as well as improved some of my existing technical skills. In addition, I have learned about the life cycle of a small start-up, structuring and tracking the progress of a dev cycle, and about the organization’s history—the idea for the company came originally from a Winter Study class on entrepreneurship!
I am so thankful to all of the folks at Whooo’s for this opportunity to learn, contribute, and collaborate this summer. Also, I would like to say thank you to Mr. and Mrs. Thomsen for sponsoring my fantastic summer of learning. Thank you as well to the ’68 Center for Career Exploration for organizing ASIP as a whole. I am so very grateful to be supported on all sides by the Williams community!