Guidelines

Every build from the Grove Lab is a prototype only. Prototypes likely should not be deployed in the classroom.

A prototype is something to try, react to, and talk about. Before a build could be used with students, it would need further review for accessibility, security, and student data.

We’re here to help you explore what’s possible using agentic AI, but there are some areas we won’t support. Anything that involves prediction markets, gambling, deep fakes, setting traps for students in assignments, or automated spam-like bots aren’t something we’re going to build. We also aren’t keen on using AI to automate teaching or student learning; however, if you’d like to build something that helps you improve the learning experience with your students, we’re always game to hear about it.

Most builds will require some time and may take a week or more based on resources, capacity, and complexity. AIG may also not have the ability to meet all of your requests.

Builds that use outside data, run a detailed simulation, or have several connected parts take longer. If you have a date in mind, the form asks for it and how firm it is.

The form asks for your best guess. If you choose optional AI assistance, Jev offers a second read from the idea you type. Four rough levels:

• Simple: one page, no outside data.
• Moderate: several screens, saved progress, and an export.
• Involved: live outside data or a detailed simulation.
• Complex: accounts, servers, live AI, or student records. These need a conversation before building.

Prototypes are built to run in the browser with nothing sent anywhere. Work saves to the visitor's own device, and they download it themselves.

If your idea needs logins, accounts, a database, or file uploads, say so in the form. Each of these raises data-security questions and usually moves a build past the prototype stage, so we'll talk about it before building.

FERPA protects student education records. Prototypes from the lab should not collect, store, or send anything that identifies a student: names, ID numbers, emails, grades, or student work tied to a person.

By default, a build saves work only in the student's own browser and lets the student download it. If students need to hand something in, they upload their downloaded file to Blackboard themselves.

Please don't include student names, grades, or work in your submission or its attachments.

Builds can pull public data from APIs, archives, and datasets so students work with real evidence. Some sources need keys, have rate limits, or carry license and terms-of-use conditions. Keys are never placed in the page itself.

The form asks which sources you have in mind, whether live data or a saved snapshot is better, and what the build should show about its evidence.

The form also asks whether the app itself should use an AI model while people use it. Live AI inside an app adds cost, keys, and data questions. Most gallery builds use no AI at all.

The ADA Title II rule for web content names WCAG 2.1 AA as the standard for public universities.

Prototypes aim for it: keyboard access, screen-reader labels, color contrast, captions and transcripts, reduced motion, and layouts that work on phones. A prototype has not been formally audited. If you plan to use a build with students, it needs review first.

The form asks about accessibility needs you already know about, and whether your build will include audio or video.

I'd like to invite you to share your build openly during a forthcoming AI Institute or a meeting of the AI taskforce.

You decide in the form: present it yourself, let AIG show it, decide later, or keep it private. You can also say whether the prototype may appear in a public gallery with credit to you.

You can opt in to complete an optional reflection about the implications and possibilities agentic AI might have for your teaching and research.

It is open ended. You can write it in the form or after you've tried your prototype, and you choose whether it is public with your name, public without your name, or private.

The first step of the build form offers optional AI assistance. You decide whether to use it. Your request is handled the same either way, and AIG reads every form in full.

What happens if you choose it

The text in the "Here's my idea" box, and only that text, is sent to Jev, a small AI model from TypeSafe AI, through a proxy that keeps the API key private. Jev returns probabilities, not verdicts: which pathway fits, where the idea sits on a readiness rubric (purpose, audience, complexity), and whether it seems to involve student data, outside data, or AI running inside the app. The form shows how sure Jev is about each. Everything Jev produces is labeled as AI-generated on the page and in your download, and you can remove it at any time.

AI assistance may help if

• you aren't sure which pathway fits your idea
• your idea mixes several kinds of builds
• you'd like a quick check of what's still unclear before you fill out the form
• you're curious to see how a small AI model reads a request

You may want to skip it if

• you already know which pathway fits
• your idea includes anything you'd rather not send to an outside AI service
• you'd rather think through the pathways yourself first
• you or your department avoid AI tools for this kind of work

If you skip it, nothing is sent anywhere, and you choose your pathway using the descriptions and gallery examples. Jev does not decide whether or how your idea gets built, and it can be wrong.

1. Fill out the build form. It saves in your browser as you type. Nothing is sent until you email it.
2. In the last section, choose Download for Word (.doc) or Download as a web page (.html). Either works.
3. Write an email to mwatkins@olemiss.edu with the subject line "Grove Lab build idea: [your working title]".
4. Attach the downloaded file, plus any readings, images, data, or rubrics you mentioned.
5. Send it. AIG will reply to talk through the build.

Prefer paper or Word? The last section of the form also has a blank form you can download, fill in, and attach the same way.