Oct 2026
WeBWrite
Homework where students answer in their own words, and Gemma checks each answer against the teacher's key ideas as they type.
- Result
- Best Use of Gemma (Google DeepMind)
- Built with
- Next.js, Prisma and PostgreSQL, Gemma 4 through the Gemini API, Vercel
Homework tools like WeBWorK can grade a number, because a number is either right or wrong. An explanation is harder: there are a dozen ways to say the right thing, and a dozen more to almost say it. WeBWrite lets teachers write questions that need an answer in words, along with the key ideas a good answer should contain.
As the student types, Gemma reads the answer and marks each key idea as missing, getting there, or included, and shows the teacher's hint for anything still missing. All of a question's key ideas, plus a check for anything factually wrong, are judged in one call that takes about a second, so the feedback keeps up with the typing. That way, we can grade essays fairly, since a unpredicted but defensible position shouldn't be as marked wrong. And multiple-choice questions are, of course, marked exactly.
The interface is in Québec French by default, with English a toggle away. As part of the team, I handled the platform's integration with Gemma. We built it at Hacktoberfest Hack Day Montréal, where it won the Best Use of Gemma prize from Google DeepMind.