MacEduStack: A Generative AI Assessment Development Tool

Fazmin Nizam is the IT Infrastructure and AI Lead for the Faculty of Engineering at McMaster University and has previously served on McMaster’s Generative AI Task Force. He has over 17 years of experience in the fields of IT infrastructure, Software Engineering, AI/ML, Virtualization, and cybersecurity in the higher education sector. Fazmin began experimenting with Generative AI prior to the release of Large Language Models (LLMs) and has continued to build his expertise in generative AI since. He has been actively designing MacEduStack, a single, unified platform that enables educators to quickly explore a wide range of assessment options for their courses.  

The New Process

The MacEduStack platform is a web-based workspace built around a simple idea: the AI drafts, and the educator reviews everything before students see it. It draws on recent AI models from several providers and uses AI tools for web searches that bring current, real-world events into its outputs. The institution chooses which AI models are used, so those decisions stay with the university rather than with individual vendors. Fazmin built the platform on a foundation of educational research concepts such as multiple taxonomies, varied methods of assessment, different types of rubrics, Universal Design for Learning principles, competency frameworks, and many more. Accessibility was treated as a core design principle from the start, so that the platform is usable by everyone, educators and students alike.

One of the most useful features of MacEduStack is its ability to let an educator develop an AI-generated assessment based on their specific course details, and then quickly rework it, for example by converting the rubric to another style or turning the assignment into alternative formats that measure the same learning. This means educators can try different approaches without starting from scratch, and spend their time on judgement and teaching rather than first drafts, all in a simple, easy-to-use interface.

Fazmin also wanted to consider students’ perspectives when designing MacEduStack. He created a student interface where students can take quizzes their educators have released and build their own practice quizzes on any topic. They may, for example, use AI-generated multiple choice quizzes to test their knowledge of course content with an explanation for each question to help them understand where they have made mistakes. They can also flag confusing questions to bring to office hours, and a reflection journal lets them reflect on their learning processes. Students choose whether to share each reflection with their instructor.

Importantly, Fazmin decided early on that generative AI would not make decisions about students’ grades, given students’ concerns about their work being marked by AI. Quizzes are marked against an answer key, AI scoring of conversations happens only when the educator turns it on, and AI summaries of reflections never change a mark. The platform is also not yet connected to the university’s learning management system, so students use it to practise for a specific course while professors still deliver and evaluate formal assessments through existing processes.

What functionality can an Educator use?

• Uploading Course Information: Educators can upload a syllabus, such as a Simple Syllabus course outline, and the system will fill in the course name, code, description, learning outcomes and evaluation details for them to check. They can also upload course documents such as textbook chapters, notes or slides, which the system summarizes into key topics and concepts. The system then uses this information to help educators expand on or refine their assessments.
• Guardrails: Educators can set course AI rules: short instructions the AI must follow whenever it writes for the course, whether the educator is creating an assessment or a student is building a practice quiz. Each rule has a category (Content, Difficulty, Format, Assessment, or Ethics and Accuracy) and a priority, and the system can suggest a starter set. Programs and faculties can also set rules that apply to every course beneath them.
• Built-in Universal Design for Learning Principles: The system draws on Universal Design for Learning (UDL) principles taken directly from the CAST UDL Guidelines. Educators can run UDL Insights, which reviews an assessment against the three UDL principles and shows where it is already inclusive and where it could offer students more choice in how they engage, take in ideas and show what they know. UDL also shapes the platform’s breadth: with dozens of traditional formats, 19 alternative formats, ten question formats and six rubric styles, giving students a different way to demonstrate their learning takes minutes rather than days.
• Accessibility by Design: Accessibility is built into the interface itself. Every page, including the sign-in page, offers four text sizes, high contrast and inverted colour modes, and text-to-speech with a choice of voice, speed and pitch. The platform also supports light and dark modes, keyboard navigation with a “Skip to main content” link, phone-friendly screens, and properly displayed mathematics and chemical formulas.
• Built-in Educational Taxonomies: When creating an advanced assessment, educators can choose from a range of educational taxonomies, including Bloom’s Taxonomy, the SOLO Taxonomy and a Creativity The chosen taxonomy shapes how the assessment and its rubric levels are described, so educators can quickly realign their work to different taxonomies based on their course learning outcomes. 
  
• Rubric Generation and Conversion: The system can generate an initial rubric for an assessment or build an assessment around an existing rubric. Educators can also quickly switch between six rubric types (analytic, developmental, holistic, single-point, scoring and checklist) to experiment with alternative options.
• MCQ (Multiple Choice Question) Generator: Using the course content and selected learning outcomes, the system can generate sets of up to 20 Multiple Choice Questions, each with an explanation, that educators can use in real assessments or provide to students for practice. Educators choose the difficulty and the number of answer options, can ask a second AI model to check each answer key, and can switch any question between ten formats (e.g., select all that apply, true/false, rank order, case-study vignette), so students have more than one way to show what they know.
• Model-based and Simulation-based MCQ: Multiple Choice Questions can also be developed using models and interactive simulations. Each simulation ties two to seven questions to a small live model with sliders, live outputs and a chart, so students answer by experimenting rather than from memory. The system first checks whether a topic suits a simulation and explains its verdict. This helps educators develop problem sets for real assessments or student practice. For numerical disciplines, question sets can also give each student their own values in the same question, so students practise the same skill but cannot simply copy each other’s answers.
• Experimental Assessments: This section of the system offers 48 AI-era assessment designs in nine families, in which AI is part of the task rather than something to keep out of it, allowing educators to quickly try out a variety of assessment types with their course content. Examples include The Audit, where students find and correct flaws in subtly wrong AI-generated work, and The Decision Log, where students justify whether to use AI at each step of a task. A guided chooser can also recommend a design based on the educator’s learning outcomes.
• Oral Assessments and Educational Pods: The system can generate a bank of up to 10 oral questions in one of four interview styles, such as a process walkthrough or a follow-up to an essay or lab report. Each question includes the expected concepts and follow-up probes, and can be played aloud so educators can check how it sounds. Oral questions also give students an alternative to written work for showing their understanding. Educators run the interviews themselves; students cannot yet record answers in the platform, pending consent, encryption and a privacy review. Educators can also build educational pods: AI conversation partners that students chat with and whose conversations educators can review afterwards. Each pod plays one of six roles: Tutor, Examiner, Socratic, Debate partner, Viva or Custom. For example, a Socratic tutor can be created that will guide a student through Socratic questioning. 

The Impact and Feedback

MacEduStack is still a work in progress. It draws on lessons learned from earlier assessment tools that Fazmin developed with the McPhearson Institute (Assessment Partner). Those tools had over 1,800 registered users who generated more than 7,000 assessments during the initial external pilot. Overall, users found the interface user-friendly and appreciated it as an easy way to generate assessment ideas.

The software also provides analytics to observe student use and student performance. For example, quiz analytics show the number of students and attempts, average and best scores, and the most and least challenging questions, and the AI can summarize the results with topics to revisit. Class insights drawn from shared reflections also highlight common strengths and struggles.

Upon feedback from some of the users, Fazmin implimented an ‘energy footprint’ tool which estimates the electricity, water and carbon emissions used by each request and translates them into everyday comparisons. For example, it can show how the energy used by a user’s prompts compares to a number of web searches or kilometres driven in a car. Users who prefer not to see it can switch it off.

Ongoing Work

MacEduStack is currently being piloted. As part of this process, the MacPherson Institute is in the process of setting up a pilot study to evaluate the platform ahead of broader deployment, and its potential use at scale within the Faculty of Engineering is also being explored. Privacy has been treated as a design requirement from the start: AI providers receive no names or email addresses, and student reflections stay private unless shared. Further steps are under way before wider use, including retention schedules, agreements with AI providers and a formal privacy impact assessment.

—

Nizam, F. (2026), MacEDUStack, A Generative AI Assessment Development Tool,  Teaching in the Age of AI: Examples, MacEduStack: A Generative AI Assessment Development Tool. Licensed under Creative Commons BY-NC-SA 4.0.