Projects / AI-X Framework
ActiveAI-X Framework: guided self‑evaluation for AI learning tools
1 / 4The home page
- Product
- The AI-X Framework, a free web app for evaluating AI tools and ideas for learning.
- Role
- Design lead
- Team
- Nishad P (design lead), with two designers
- LEI's engineering team of 3
- the framework by 3 of LEI's researchers
- Timeline
- October 2025 onward
- Status
- Live and free to use, and I still lead its design
- new AI features designed
- an MCP version in progress
- Made for
- ASU's Learning Engineering Institute (LEI), as part of a research project funded by the Kern Family Foundation and the Institute of Education Sciences (IES) Active Learning at Scale grant
- My tools
- Figma
- Tailwind with Flowbite
- CSS and SCSS
- AI coding agents
- The app runs on
- Angular, with Google and Microsoft sign-in, on LEI's AWS
AI-X helps educators and designers judge an AI tool or idea for learning, and decide what to do next.
People are guided to the evaluation that fits the stage they're at. They answer reflective questions, rate their own answers, and get a result with next steps.
It's a reflective instrument, not a pass-or-fail score.
Before October 2025, the framework existed only as LEI's research documents, with no tool or interface.
The questions and the scoring come from that research. I lead the design of how people meet them: the guided start, the evaluations and their results, and the design system underneath.
AI-X is the first project where I ran my own design team and assigned the work, and it's where we created our Tailwind design system.
Impact
184evaluations
In use
From its first test in January 2026 to the end of July, 105 people ran 184 evaluations of their own AI tools and ideas, across 124 projects. 126 of them went all the way to a result.
43people
People come back
43 people ran more than one evaluation, and 39 used both the Compass and the Guide.
about44,000words
Real reflection
In the Compass alone, people wrote 993 answers, about 44,000 words on their own ideas.
41faculty and staff
Shaped by the people who use it
41 faculty and staff from 18 ASU colleges and units worked through the app in monthly sessions across Spring 2026, and the design changed after them, from how projects are found to how people leave an evaluation safely.
55instructional designers
Run in LEI's workshops
In a four-part series with ASU's AI Acceleration team, participants evaluated an AI tool with the Guide and produced a mock report. A workshop for 55 instructional designers followed.
LEI's framework became a tool people use
What was a set of research documents is now a free web app, where educators and designers evaluate their own AI tools and ideas and leave with next steps.
Research behind the framework, by LEI's researchers
A design system LEI builds on
The ASU version of Tailwind and Flowbite that my team and I created for AI-X became the base for LEI's next product, ReQUESTA.
Guided to the right tool
People shouldn't have to work out which evaluation fits them, so I designed the start to do it for them.
The home page asks what they want to do: shape an idea, check a prototype, or build responsibly.
It then suggests the tool for that stage: the Compass, the Guide or the Toolkit.
Once signed in, a dashboard keeps every evaluation in a project folder, with its status, so people can resume or revisit it.
Two evaluations and a library
The Compass, for an idea.
People name the evaluation and work through five categories, from the use case to feasibility and responsibility.
For each question they write an answer, then rate how clear and complete it is.
The result places the idea in one of four quadrants, from Set Aside for Now to Build and Implement Thoughtfully, by its likely impact on learning and how feasible it is to build.
The Guide, for a product or prototype.
People pick the themes they want to evaluate, out of eight, such as human-centeredness or technical sustainability.
For each criterion they read its definition and indicators, rate how far it's implemented, from Implemented Intentionally to Not Applicable, and leave notes.
The results sort the themes into strong practices and areas for continued development, each with questions to consider, and the report exports as a PDF.
The Toolkit, to explore or to follow from a result.
The Toolkit is a library of techniques, such as Five Whys and task analysis, arranged by the four stages of building a learning tool: understanding the challenge, creating the tool, implementing it, and investigating how well it works.
People can open it and explore at any time, and both evaluations end in next steps that point into it.
How we got there
I conceptualized the screens, the workflows and the overall UI, set the direction, verify all the design, and govern how the design system and its code are written.
It started with my paper sketches of the landing page, the dashboard, each evaluation and its results. From those I led the wireframes and the high-fidelity design with my team. In my first designs, the evaluations show one question at a time, section by section, so a long evaluation doesn't overwhelm people. The tools' names changed several times on the way, as the research sessions shaped them. I also made the Compass explainer video, hosted on the AI-X site.
What testing with faculty changed
LEI's researchers tested AI-X in rounds.
First, 30 faculty and staff reviewed the framework on paper, in four focus groups in Fall 2025, before there was an app.
The live app's first test came in January 2026, and at its launch at ASU's FOLC Fest in February, about 25 people used it at once, 5 of them on phones.
From January to May, 41 faculty and staff worked through it in five monthly sessions and talked through what got in their way.
The sessions doubled as usability testing, and we reworked the app after them:
Projects firstcreating a project was hard to find, so now every evaluation belongs to one, and the dashboard opens the first folder, so finished work is easy to find.
Themes are opt‑inin the Guide, people choose themes rather than remove them, and selected themes are clearly marked.
Leaving is safeBack became Exit evaluation, with a warning, and the second back button went.
Clearer words"Save for later" became "Mark for review".
Easier to tell aparteach tool got its own icon, and FAQ and About moved into the top bar.
Five landing pages, comparedI designed five versions of the landing page, and faculty in five breakout groups compared them to shape the one we built.
What people wrote after the first test, in January 2026:
“The interface of the framework is well designed as well. Good work!”
“Easy to use! I like that you only see one question at a time rather than seeing a full list of qustions you need to answer.”
“The tech part was smooth. Interface is clear, and instructions are easy to follow.”
“It's very clean and well organized”
The design system
AI-X is where we realized that LEI, building more ASU-facing tools, needed a design system for that wider scope. I was asked to find a reusable system.
I chose a Tailwind kit with Flowbite, and with one of my designers made it ASU's, following ASU's brand guidelines. I implemented it in code, in CSS and SCSS, and my designer filled in the details.
The token layer AI-X runs on in production, the colours, type and spacing every screen draws from, is the stylesheet I wrote.
The code itself is generic, but writing it taught me how to arrange the classes and what granularity each level needs, and it showed me the difference in philosophy between a utility-first framework like Tailwind and a component-based one like Bootstrap. The same system carried on to ReQUESTA.
What's being added
LEI's original spec for the app left AI for later.
Now my team and I have designed three AI features for the Compass:
Copy for AIcopy your answers, or open them straight in Claude or ChatGPT, to keep thinking them through with an AI assistant.
AI tipshints personalized from your earlier answers.
A personalized resulta reading drawn from your own evaluation.
Next, we're working on a version that runs through MCP, the standard way AI assistants connect to tools.
It will serve the questions dynamically, and evaluate the answers as they come.
LEI's proposal for this next phase names me to lead its UX design and oversee prototyping and usability testing.
Reflection
AI-X is shared work, and it's hard to untangle. The framework isn't mine: it comes from research and many studies. But people meet it through the UI, and the UI is mine, so when someone gets stuck, it's often unclear whether it was the wording of a question, the UX copy, the interface, or how the questionnaire and its results are structured.
It's also the first project where I assigned the work to a team, and leading means I'm responsible for every output. I set the direction, verify all the design, and govern how the design system, the CSS and SCSS, and the high-fidelity design are made, while my team does 60% or more of the hands-on work.
I can't claim it alone, and with everything else I work on, I couldn't have done it alone.
A question I hold on my other projects comes up here too: where people should reach an evaluation at all, whether through this UI, an AI assistant, Canvas (ASU's learning management system), an extension or something else. As we open the framework to those surfaces through MCP, the web UI's perceived usability drops for stakeholders, and what they expect from it changes.