Hey Dr. Kachorsky — the UDL framing is the right move here. Most AI-in-education content starts with the tool and works backward to the pedagogy. You started with the learner and worked forward. That's why this lands differently.
The Q&A chatbot section is the one I want to push on because I think you're underselling it. I build custom AI agents and the guardrails piece — "answers clarification questions but can't complete work" — that's not just a setting you toggle. That's a design decision that changes everything about how the student interacts with it. A well-scoped agent that knows the assignment, knows the rubric, and knows where the line is between helping and doing? That's a fundamentally different tool than ChatGPT with instructions stapled on.
The reading level adaptation point is huge for parents too, not just teachers. I've seen agents built for tutoring that adjust in real time based on how the student responds — not just set to "7th grade" but actually meeting the kid where they are in the conversation. That's where this goes next.
Also — love that you were transparent about the audio-to-AI-to-human workflow. That's modeling exactly what you're teaching. More educators should see that.
Hi Colleen. Thank you for reading and thank you for your comment. I agree with you about the power potential with AI chat bots. Creating specialized bots for specific tasks or assignments is something we regularly do on our campus. We've found a lot of success with Flint, which is a platform specialized for educational purposes that does a nice job of walking users through the design process. But, putting those parameters and controls in place is critical. Of course, kids can go outside of the bots that we create, but I've found that when that happens, it just opens up an opportunity to have a really candid discussion about the process of learning and how productive struggle remains important. We actually have a great podcast episode where one of our students talks about how he created his own bots to specifically help him improve in an AP class without taking away his own learning processes that I think you might enjoy: https://mindfulaiedu.substack.com/p/mindful-ai-for-education-how-a-good
Dani — this is exactly the kind of response I was hoping for. The fact that you're already building specialized bots per assignment and framing the "going outside the bot" moment as a teaching opportunity instead of a compliance failure? That's the whole ballgame. Most schools treat AI boundaries as walls. You're treating them as conversation starters. Huge difference.
Listening to that podcast episode now — a student building his own bots to improve in AP without shortcutting his own learning is the exact use case that should be in every education AI presentation. That's not a student cheating.
That's a student who understood the tool well enough to scope it himself. That's the skill.
This is great — clear, practical, and very actionable.
I’m not in education, but these principles feel broadly applicable across other domains as well. The idea of using AI to reframe, scaffold, and personalize learning translates well into enterprise settings.
I’m curious — do you have any validation frameworks in place? For example, when AI rewrites or reframes content, how do you evaluate accuracy, fidelity to the original intent, or unintended distortions?
In professional environments, especially, the validation layer becomes as important as the generation layer.
Would love to hear how you’re thinking about that.
Great question. I don't have a specific, formalized validation process beyond reviewing the material. It is important to know the original source material well enough to compare it to rewritten or reframed content. I do not recommend sharing rewritten/reframed content with students without reviewing it to ensure accuracy and intent are maintained.
For example, I recently had NotebookLM create a presentation from an article on the methods section of a research paper to share with my AP Research classes. The presentation had a few minor mistakes and used a metaphor to convey information that was convoluted. I decided not to use the presentation and instead had NotebookLM create an outline for a presentation that I edited, then fed to Gamma to produce a more accurate (non-metaphorical) presentation. Gamma offered more control and allowed me to edit the presentation in a way that NotebookLM doesn't currently permit. To do this, I obviously needed to know the original source material, review the presentation, and make adjustments.
Similarly, some of the early assignment instruction videos I created with NotebookLM overemphasized certain aspects of the assignments. While this wasn't inaccurate per se, it was misleading. I think students would have spent more time on aspects of the assignments that were not relevant to the evaluation/grading. Without reviewing the videos, I would not have known that the videos were emphasizing the wrong things. Now, instead of feeding the AI my full assignment description, I have ChatGPT or Claude simplify the instructions first, edit out redundancy--which I suspect is what leads to overemphasis in NotebookLM, though I can't confirm that--and then, share the simplified version with NotebookLM to adapt into video form. When I review the videos now, I don't see that same overemphasis. It takes an extra step, but it is still worth it to provide students with the additional resource.
Like I said, not a "formal" framework for reviewing the adapted material, but what to me seems to be a natural process in education. I don't know of any teacher who would put something in front of their students that they hadn't reviewed first for accuracy, intent, and alignment with curricular/instructional goals. So, maybe these three things are the start of one...
This is great! I’m really interested in the Q&A chatbot for assessments - that could be really useful for my students (and me), definitely something I’ll think about doing. Thanks for sharing your ideas!
This is a masterclass in practical AI for education. I love how each strategy centers accessibility while benefiting all learners. Rewriting, alternative formats, step-by-step videos, and chatbots, a blueprint for making learning both inclusive and scalable.
For more AI trends and practical insights, check out my Substack where I break down the latest in AI.
The Q&A chatbot idea with guardrails is the one I'd highlight most because it solves a real structural problem: students need support at 11pm, teachers don't
The guardrail design is everything though. "Can explain, can't produce" is a meaningful boundary, but it requires careful prompt engineering to hold consistently
Worth testing edge cases before deploying; students are remarkably creative at finding workarounds.
One thing I'd add to your list: AI can help document accommodations more efficiently too
Generating draft IEP-aligned lesson modifications from a base lesson plan cuts prep time significantly and keeps differentiation consistent across units
Thanks for sharing what's actually working in a real classroom; this is more useful than most of what gets written about AI in education.
Making Lessons Truly Inclusive for Diverse Learners
I would like to share an observation from my experience working in schools and invite discussion from colleagues about how we can better support diverse learners in mainstream classrooms.
In many well-functioning public schools in NSW, Australia, there is a small but dedicated group of teachers often referred to as Diversity Teachers. Their role is not only to support individual students, but also to help classroom teachers design lessons and assessments that are genuinely inclusive.
I had the opportunity to work as a Diversity Teacher in Sydney. One of my core responsibilities was to collaborate with subject teachers across different departments. Together we would review lessons and identify ways to include students who may be neurodiverse, learning English as an additional language, or adjusting to a new education system.
In theory, many schools follow the Universal Design for Learning (UDL) model, which encourages teachers to plan lessons that accommodate a wide range of learners from the start. However, in practice this can sometimes become a hit-and-miss exercise unless there is someone with specific expertise guiding the process.
Classroom teachers already manage heavy workloads: lesson preparation, assessment marking, behaviour management, pastoral care, and administrative responsibilities. Expecting every teacher to become an expert in neurodiversity, language acquisition, and inclusive assessment design is often unrealistic without support.
My thrust was to add a mini project based STEAM learning in all subjects. Teachers were not keen to include hands on learning experiences due to preparation and non availability of resources as well as the compliance pressures of 'covering the curriculum bullet points'.
This is where dedicated diversity teachers can make a significant difference.
For example, part of my role involved helping teachers prepare small but meaningful adaptations within lessons. These could include:
providing simplified instructions alongside the main task
creating visual supports or step-by-step guides
designing alternative entry points for learners who need more scaffolding
including hands-on or collaborative activities for students who struggle with purely written work
We also worked closely with teachers when preparing formal assessments. Sometimes small adjustments made a large difference to student access and performance. These adjustments could include:
ensuring questions were written in clear and simple language
reducing unnecessary linguistic complexity
providing readers or writers during formal tests when required
offering alternative formats that still assessed the same learning outcomes
The goal was never to reduce academic expectations. Rather, it was to ensure that the assessment measured the intended learning, rather than a student’s difficulty in decoding complex language or navigating unfamiliar task formats.
Another important aspect of the diversity teacher’s role was acting as a bridge between specialist support services, classroom teachers, and families. Many teachers appreciated having someone who could advise on practical strategies and share insights about students’ learning needs.
From my experience, when this support structure is present, several positive outcomes occur:
teachers feel more confident designing inclusive lessons
students who might otherwise disengage remain involved in learning
assessments become more equitable and accessible
classrooms become more collaborative and supportive learning environments
Without this support, even well-intentioned inclusive models can struggle to be implemented consistently.
I am interested to hear from colleagues in other regions and education systems:
Do your schools have dedicated diversity or inclusion teachers?
How are classroom teachers supported in designing lessons for neurodiverse learners?
Is Universal Design for Learning working effectively in your context, or does it need stronger specialist oversight?
What practical strategies have worked well in your classrooms?
Inclusive education is widely discussed in policy, but its success often depends on the practical structures schools put in place to support teachers and learners.
It would be valuable to hear different perspectives and experiences.
The guardrails in #4 are the most important part of this whole piece! And probably the part most people would skip ~
Defining what the chatbot can't do is harder and more valuable than defining what it can do. That boundary is what makes it a learning tool instead of a shortcut.
The UDL connection is spot on too. Every time I've seen someone design AI access for learners at the margins, the result works better for everyone. That's not a side effect - it's the principle working as intended.
Really appreciate the transparency statement at the end. More of this.
Hey Dr. Kachorsky — the UDL framing is the right move here. Most AI-in-education content starts with the tool and works backward to the pedagogy. You started with the learner and worked forward. That's why this lands differently.
The Q&A chatbot section is the one I want to push on because I think you're underselling it. I build custom AI agents and the guardrails piece — "answers clarification questions but can't complete work" — that's not just a setting you toggle. That's a design decision that changes everything about how the student interacts with it. A well-scoped agent that knows the assignment, knows the rubric, and knows where the line is between helping and doing? That's a fundamentally different tool than ChatGPT with instructions stapled on.
The reading level adaptation point is huge for parents too, not just teachers. I've seen agents built for tutoring that adjust in real time based on how the student responds — not just set to "7th grade" but actually meeting the kid where they are in the conversation. That's where this goes next.
Also — love that you were transparent about the audio-to-AI-to-human workflow. That's modeling exactly what you're teaching. More educators should see that.
Hi Colleen. Thank you for reading and thank you for your comment. I agree with you about the power potential with AI chat bots. Creating specialized bots for specific tasks or assignments is something we regularly do on our campus. We've found a lot of success with Flint, which is a platform specialized for educational purposes that does a nice job of walking users through the design process. But, putting those parameters and controls in place is critical. Of course, kids can go outside of the bots that we create, but I've found that when that happens, it just opens up an opportunity to have a really candid discussion about the process of learning and how productive struggle remains important. We actually have a great podcast episode where one of our students talks about how he created his own bots to specifically help him improve in an AP class without taking away his own learning processes that I think you might enjoy: https://mindfulaiedu.substack.com/p/mindful-ai-for-education-how-a-good
Dani — this is exactly the kind of response I was hoping for. The fact that you're already building specialized bots per assignment and framing the "going outside the bot" moment as a teaching opportunity instead of a compliance failure? That's the whole ballgame. Most schools treat AI boundaries as walls. You're treating them as conversation starters. Huge difference.
Listening to that podcast episode now — a student building his own bots to improve in AP without shortcutting his own learning is the exact use case that should be in every education AI presentation. That's not a student cheating.
That's a student who understood the tool well enough to scope it himself. That's the skill.
Really glad to have found this publication.
This is great — clear, practical, and very actionable.
I’m not in education, but these principles feel broadly applicable across other domains as well. The idea of using AI to reframe, scaffold, and personalize learning translates well into enterprise settings.
I’m curious — do you have any validation frameworks in place? For example, when AI rewrites or reframes content, how do you evaluate accuracy, fidelity to the original intent, or unintended distortions?
In professional environments, especially, the validation layer becomes as important as the generation layer.
Would love to hear how you’re thinking about that.
Great question. I don't have a specific, formalized validation process beyond reviewing the material. It is important to know the original source material well enough to compare it to rewritten or reframed content. I do not recommend sharing rewritten/reframed content with students without reviewing it to ensure accuracy and intent are maintained.
For example, I recently had NotebookLM create a presentation from an article on the methods section of a research paper to share with my AP Research classes. The presentation had a few minor mistakes and used a metaphor to convey information that was convoluted. I decided not to use the presentation and instead had NotebookLM create an outline for a presentation that I edited, then fed to Gamma to produce a more accurate (non-metaphorical) presentation. Gamma offered more control and allowed me to edit the presentation in a way that NotebookLM doesn't currently permit. To do this, I obviously needed to know the original source material, review the presentation, and make adjustments.
Similarly, some of the early assignment instruction videos I created with NotebookLM overemphasized certain aspects of the assignments. While this wasn't inaccurate per se, it was misleading. I think students would have spent more time on aspects of the assignments that were not relevant to the evaluation/grading. Without reviewing the videos, I would not have known that the videos were emphasizing the wrong things. Now, instead of feeding the AI my full assignment description, I have ChatGPT or Claude simplify the instructions first, edit out redundancy--which I suspect is what leads to overemphasis in NotebookLM, though I can't confirm that--and then, share the simplified version with NotebookLM to adapt into video form. When I review the videos now, I don't see that same overemphasis. It takes an extra step, but it is still worth it to provide students with the additional resource.
Like I said, not a "formal" framework for reviewing the adapted material, but what to me seems to be a natural process in education. I don't know of any teacher who would put something in front of their students that they hadn't reviewed first for accuracy, intent, and alignment with curricular/instructional goals. So, maybe these three things are the start of one...
Thank you. Appreciate the detailed response.
This is great! I’m really interested in the Q&A chatbot for assessments - that could be really useful for my students (and me), definitely something I’ll think about doing. Thanks for sharing your ideas!
This is very interesting and echos something Im working on currently. The future is recognising when and how to use Ai!
outstanding post. so good
Hi, completely relate. I write about humanizing the future of learning.
https://substack.com/@devikatoprani/note/p-177581013
This is a masterclass in practical AI for education. I love how each strategy centers accessibility while benefiting all learners. Rewriting, alternative formats, step-by-step videos, and chatbots, a blueprint for making learning both inclusive and scalable.
For more AI trends and practical insights, check out my Substack where I break down the latest in AI.
this substack is excellent for educational research on AI and LLMs. Very accessible.
https://theharness1.substack.com/p/how-a-language-model-actually-works?r=fw5j9&utm_medium=ios&shareImageVariant=solid
interesting article
Loved this
That is an awesome article for teachers using AI, thanks!
The Q&A chatbot idea with guardrails is the one I'd highlight most because it solves a real structural problem: students need support at 11pm, teachers don't
The guardrail design is everything though. "Can explain, can't produce" is a meaningful boundary, but it requires careful prompt engineering to hold consistently
Worth testing edge cases before deploying; students are remarkably creative at finding workarounds.
One thing I'd add to your list: AI can help document accommodations more efficiently too
Generating draft IEP-aligned lesson modifications from a base lesson plan cuts prep time significantly and keeps differentiation consistent across units
Thanks for sharing what's actually working in a real classroom; this is more useful than most of what gets written about AI in education.
Great!
Making Lessons Truly Inclusive for Diverse Learners
I would like to share an observation from my experience working in schools and invite discussion from colleagues about how we can better support diverse learners in mainstream classrooms.
In many well-functioning public schools in NSW, Australia, there is a small but dedicated group of teachers often referred to as Diversity Teachers. Their role is not only to support individual students, but also to help classroom teachers design lessons and assessments that are genuinely inclusive.
I had the opportunity to work as a Diversity Teacher in Sydney. One of my core responsibilities was to collaborate with subject teachers across different departments. Together we would review lessons and identify ways to include students who may be neurodiverse, learning English as an additional language, or adjusting to a new education system.
In theory, many schools follow the Universal Design for Learning (UDL) model, which encourages teachers to plan lessons that accommodate a wide range of learners from the start. However, in practice this can sometimes become a hit-and-miss exercise unless there is someone with specific expertise guiding the process.
Classroom teachers already manage heavy workloads: lesson preparation, assessment marking, behaviour management, pastoral care, and administrative responsibilities. Expecting every teacher to become an expert in neurodiversity, language acquisition, and inclusive assessment design is often unrealistic without support.
My thrust was to add a mini project based STEAM learning in all subjects. Teachers were not keen to include hands on learning experiences due to preparation and non availability of resources as well as the compliance pressures of 'covering the curriculum bullet points'.
This is where dedicated diversity teachers can make a significant difference.
For example, part of my role involved helping teachers prepare small but meaningful adaptations within lessons. These could include:
providing simplified instructions alongside the main task
creating visual supports or step-by-step guides
designing alternative entry points for learners who need more scaffolding
including hands-on or collaborative activities for students who struggle with purely written work
We also worked closely with teachers when preparing formal assessments. Sometimes small adjustments made a large difference to student access and performance. These adjustments could include:
ensuring questions were written in clear and simple language
reducing unnecessary linguistic complexity
providing readers or writers during formal tests when required
offering alternative formats that still assessed the same learning outcomes
The goal was never to reduce academic expectations. Rather, it was to ensure that the assessment measured the intended learning, rather than a student’s difficulty in decoding complex language or navigating unfamiliar task formats.
Another important aspect of the diversity teacher’s role was acting as a bridge between specialist support services, classroom teachers, and families. Many teachers appreciated having someone who could advise on practical strategies and share insights about students’ learning needs.
From my experience, when this support structure is present, several positive outcomes occur:
teachers feel more confident designing inclusive lessons
students who might otherwise disengage remain involved in learning
assessments become more equitable and accessible
classrooms become more collaborative and supportive learning environments
Without this support, even well-intentioned inclusive models can struggle to be implemented consistently.
I am interested to hear from colleagues in other regions and education systems:
Do your schools have dedicated diversity or inclusion teachers?
How are classroom teachers supported in designing lessons for neurodiverse learners?
Is Universal Design for Learning working effectively in your context, or does it need stronger specialist oversight?
What practical strategies have worked well in your classrooms?
Inclusive education is widely discussed in policy, but its success often depends on the practical structures schools put in place to support teachers and learners.
It would be valuable to hear different perspectives and experiences.
!!
The guardrails in #4 are the most important part of this whole piece! And probably the part most people would skip ~
Defining what the chatbot can't do is harder and more valuable than defining what it can do. That boundary is what makes it a learning tool instead of a shortcut.
The UDL connection is spot on too. Every time I've seen someone design AI access for learners at the margins, the result works better for everyone. That's not a side effect - it's the principle working as intended.
Really appreciate the transparency statement at the end. More of this.