Automated Worksheets for K-12th Teachers with MateriALL

Using NLP technologies to auto-generate classroom materials for teachers

A high-fidelity modal I designed for the MateriALL Google Add-on


K-12th teachers do not have enough time to do their jobs. A significant portion of this time is spent making classroom materials. Many of these materials are minor variations of one another — for example, a teacher might design a worksheet based off of a previously designed slide deck.

MateriALL is a Google Add-on that takes curriculum content that already exists, like a slide deck, and uses NLP technologies to generate new materials like worksheets. MateriALL was researched, designed, and built by myself along with group of two UX researchers and a software engineer for our Masters degree final capstone project.

Defining the Problem

Research for MateriALL began with qualitative interviews and surveys conducted over teachers of K-12th classrooms in the United States. The following findings ended up being the most impactful in properly scoping our product:

Younger, less-experienced teachers spend a lot of time creating materials because they don’t have existing content to use and must make everything from scratch.
Teachers mostly use Google Drive for lesson planning. For example, they might first use Google Slides to make their lecture, then Google Docs to make their worksheet.

These research findings led us to the following “How Might We” question:

How might we use the tools teachers already use, like Google Docs, to make the lesson planning process, especially for creating materials, more streamlined?

In brainstorming answers to this question, we arrived at an idea that would eventually become MateriALL - something that would take teachers’ existing slide content, like definitions and facts, then use them to automatically generate and/or format worksheet questions. With these insights, we dove into the research, design, and development of MateriALL.

Defining User Control feat. all Stakeholders

One of my mid-fidelity screens used for concept-testing. Only certain question types, like true/false ones, are available based on selected content.

We defined some key functionalities for MateriALL by considering both our team's skillsets and insights from our initial user interviews. However, as we would soon find out, our decisions were rife with unspoken assumptions.

Only certain slide content could create certain types of questions. We assumed that curriculum content was inflexible and could only “fit” one type of worksheet question.
Users could only either use the automatically-generated question that MateriALL provided, or write their own. We assumed that our NLP technology could choose an “optimal” question statement.
I iterated on the prior screen -- now, users can select any content and generate any questions with it.

When we concept-tested these functionalities with users using mid-fidelity wireframes, we found that teachers were uncomfortable with MateriALL's rigidity. Teachers wanted more control and flexibility because they knew what kind of content would benefit their students the best. In response, we redefined these functionalities.

Users can make any questions now out of any of the content they chose. In some cases, this might yield strange results from our NLP algorithms, but it will err on the side of flexibility.
Users can see and select from more options or write their own. What our NLP algorithms consider suboptimal might be just what our users are looking for, or spark inspiration to write other content.

By involving not just the entire product team, but also our users via concept-testing, we iterated on our initial interactions collaboratively to be more user-friendly for teachers.

Visual Design Language feat. Me

As we moved further along in the design journey, it became time to add more visual design into MateriALL. I began this process by experimenting with different color palettes, components, and even existing style guides.

Trying out different treatments helped me make informed decisions about which visual direction to go. Ultimately, I chose a design system that evoked playfulness, harkening back to grade school learning, as well as a sense of lightweight airiness, to convey the efficiency MateriALL would bring to K-12th teachers.

I tried various visual design patterns to help me visualize what would[n't] work.
Some of the high-fidelity screens I designed for MateriALL. I used Figma design components to keep things consistent.

The Outcome

Thanks to the collaborative efforts of our entire team, we presented our work and a demo of MateriALL to a panel of judges at the end of the school year. Following this...

We were awarded the prestigious Dr. James R. Chen award for our outstanding project. The judges were impressed that we “allowed [ourselves] to be guided by what [we] learned from [our] research.”
We submitted MateriALL to Google for approval, so that it could be accessible to all K-12th teachers as a Google Docs Add-on.


This was one of my largest, most hands-off design efforts yet, and I learned a lot from the experience.

Parallelizing Work
Without a lot of time, parallelizing engineering and design might be necessary. However, it's still important to squeeze in validation where we can! 
Total Stakeholder Involvement
By involving all important parties, from the builders to the users, we can yield more fruitful insights to inform design decisions.
Being THE Designer
As a solo designer, I learned how methods like visualizing options can help me make decisions on my own, when necessary.