From Compliance to Connection

I built the design on Positionality Memo 1, Critical Racial Experiences, in EDU 703, a foundational course in my own doctoral program at American University. The one-page, first-person memo is worth 200 points, and it asks students to describe critical racial moments from their own schooling and account for how those moments shape the way they later read data. The prompt asks for the culture of racism a student lived inside rather than the systems around it. Three objectives sit under it, which are reflexive awareness, culture-of-racism analysis, and lens-to-inquiry. Though the assignment is difficult on purpose, the difficulty I set out to address was the distance between a student and the first sentence.

Ryan and Deci (2000, 2017) identify autonomy, competence, and relatedness as basic psychological needs. They also name a felt lack of competence, along with a lack of interest or value, as sources of amotivation. That framing pointed my attention at the entry point rather than the work itself, since a student who cannot find a way in has no way to show me what they can do. Autonomy here comes from starting with the student's own inquiry, and competence comes from keeping the objectives visible while the student writes, but I do not claim that a generated prompt can create belonging. Relatedness is the need my design leaves alone.

The decision that shaped everything else was my decision about what the model may not touch. The objectives stay fixed, the rubric stays fixed, the memo a student turns in is the same memo every other student turns in, and what moves is only the route in. The workspace asks four short questions about the settings that shaped the student's racial experience, one critical moment from their schooling, their current inquiry, and the lens it built. From those four answers the model writes questions that use the student's own details to push them to explain, connect, and examine the thinking they arrived with. I instructed it to invent nothing, to name only what the student supplied, and to never infer identity. A model asked to personalize past what it was given will start filling the gap with what it assumes about the student.

I built the workspace as a Next.js application and deployed it at edu703memo.vercel.app, where it runs with no login, no database, and no analytics. Students type about their own racial experiences, so my architecture stores nothing, although their answers do pass through a commercial model provider once during generation, which I disclosed before use. The last step hands the student a Word document carrying the personalized prompt, their own intake, the unchanged rubric, and a place to write. The model is not called again, so it designs the entry point and the student does the writing.