BE TRANSFORMATIVE.

Augment human capability.Amplify human talent.Elevate human judgement.

This is the future.

Coursework

Assignment 1 (EDFD 70303)
AI and Cognitive AtrophyThis paper explores existing literature related to my personal theory that AI affects learners at different levels of Bloom’s Taxonomy in different ways, and we must support beneficial “scaffold” behavior instead of a negative cognitive “crutch.”The assignment asked me to engage with peer-reviewed literature to interrogate a theory I'd developed through my own professional observations.Using original research methodology to evaluate real studies and build an evidence-based argument, this paper ended up demonstrating research literacy, reflective practice, and the ethical framing of AI integration simultaneously. The scaffold vs. crutch distinction I developed here runs through almost everything else in my portfolio.Assignment 2 (LSTE 7329)
Virtual Learning Interventions: From AI to ZPDThis annotated bibliography collects and documents relevant sources of research on how AI can be used to modify a learner’s zone of proximal development (Vygotsky) and act as a powerful scaffold for teaching new information and skills.Building this annotated bibliography required me to identify, evaluate, and synthesize sixteen sources through a single theoretical lens. That process is inherently a research methodology exercise.Because every source was selected for what it could tell me about AI as an instructional scaffold, the bibliography doubles as a principled argument for how these tools should be used in real learning environments ... drawing on theory to make judgments about practice, which is exactly what Standards 2 and 3 ask for.Assignment 3 (EDFN 7314)
Philosophy of Practice for ITSThis document allowed me to convey my personal philosophy of practice for ITS, focusing on adaptive personalization, cognitive load management, feedback and reinforcement, learner agency and autonomy, socially situated learning, and collaborative interaction.Writing a philosophy of practice forced me to do something harder than summarizing research. I had to commit to a set of principles and actually defend them.The five pillars I developed (adaptive personalization, cognitive load management, feedback and reinforcement, learner agency, and socially situated learning) are each grounded in specific empirical frameworks. That grounding satisfies the research theory requirement, while the act of articulating them as a coherent professional stance is what makes this a genuine reflection on practice.Of everything in this portfolio, this document is the clearest statement of who I am as an instructional designer.Assignment 4 (EDFN 7370)
Test Analysis: Humanity’s Last ExamThis review analyzes the Center for AI Safety’s “Humanity’s Last Exam” test for assessing AI intelligence and capabilities. While this is an important early benchmark, I suspect even our best efforts to assess AI in a human-centric and human-legible “test” format are still in their infancy.I chose to analyze HLE rather than a conventional educational instrument because it sits at the exact intersection of assessment design and AI capability measurement that defines my professional focus.Applying formal psychometric criteria (i.e. validity, reliability, technical documentation, scoring utility) to a non-traditional benchmark required stretching those frameworks in ways a standard test analysis wouldn't. The assignment is an argument that rigorous assessment design matters even when the subject being assessed "isn't human."Assignment 5 (LSTE 7303)
ITS: Skinner’s Teaching Machine, EvolvedOne of my earliest papers on Intelligent Tutoring Systems, this piece traces the lineage of new AI tools back to their roots in B.F. Skinner’s work on teaching machines and explores the near horizon of Adaptive Intelligence Tutoring Systems (AITS) that adjust to learner feedback.Looking back on this paper, it laid the intellectual foundation for many of my following assignments as well as my future work. Tracing ITS development from Skinner's behavioral teaching machine through PLATO to modern adaptive systems and then forward to AITS required evaluating each generation of technology against the learning theory of its time.The conclusion I reached, that constructivism ultimately produced better tools than behaviorism, is an observation I've carried forward.Assignment 6 (LSTE 71703)
Mobile Application Planning Document: ALICEThis paper established the need for ALICE, an early adaptive reading system concept that I eventually independently developed into ALIENS (Adaptive Literacy Instruction for English and Spanish), using AI for development after I did not receive an institutional grant.That overall trajectory—from a formal design document to an independent professional project—is probably some of the clearest demonstration I have of applying what I've learned here to real work in the field.Assignment 7 (RHET 5317)
ReKedit: An AI Writing Assistant Inspired by Draft No. 4When the LSTE program was reduced by one class from 36 to 33 hours, I enrolled in a creative writing class with departmental approval to use it as an elective related to my major. This piece establishes how AI can act as a scaffolded tool that respects user agency and creative control.Reading McPhee's description of how Howard Strauss's Kedit macros transformed his structural writing process made me think immediately about what LLMs could do with the same problem.Building ReKedit turned the essay into a working design artifact. The ethical constraint I built into the tool (it cannot affect the writer's prose while helping as an early-stage editor) connects to some of my other research into cognitive atrophy.Assignment 8 (LSTE 7313)
That Cursed MuseThis multimedia project represents a “meta” analysis of controversial production technology that influenced my independent SPECTRUM+ AI literacy program for the artists. I used OpenAI’s Dall-E image generation model to produce a gallery of historically disruptive media.I published the exact prompt I used because transparency about the tool is part of the viewer experience. The nine historical examples build the argument that humanity's resistance to new creative tools is cyclical. This assignment gave me the confidence to create the SPECTRUM+ AI Literacy for Artists framework in my Projects section.Assignment 9 (LSTE 7304)
Chat Masterclass: Visible vs Invisible UX in ChatGPTThis video lesson teaches a unique and powerful UX for engaging with multiple AI agents in a single conversational workflow. I used the OBS screen recording program and edited the final video in Canva. This methodology is an emerging best practice in conversation design, or CXD.Most people interact with ChatGPT exclusively through the visible interface (the chat window) without ever engaging the system prompt or custom instruction layers where the tool's real power lives.I think this multi-agent configuration represents an emerging professional competency that I hope to share with my co-workers after graduation. In retrospect, I want to improve my narrative performance and sound editing on video lessons like these so that it improves the overall learner experience.Assignment 10 (LSTE 71703)
MAP Group Project Part 1: Write-WiseFor this group mobile learning project, I led design and UX development for the Write-Wise app. The project idea originated from another member of the team, and I helped translate the written description into storyboarding visuals with their inclusive feedback.I was making the design decisions that determined how the app would actually feel to use. Working from someone else's idea and making it visually and structurally real while keeping it a "team effort" is the kind of collaborative learning experience that makes me appreciate group curriculum and LXD work.

Projects

While enrolled at the University of Arkansas at Little Rock, I created a series of independent extracurricular projects related to AI in human learning and workforce development. These applications were developed outside the scope of my institutional education (despite some attempts to formalize the work early on), and I used AI as a tool under my instruction in a human-guided design process to create the final versions that I have shared below.I believe these are valuable contributions to the educational community.Unless otherwise noted in a separate licensing agreement issued under my explicit consent, I maintain sole copyright and IP owner status of the works and content described on this site or hosted elsewhere. Some of my projects have been openly released under an MIT license in accordance with the best practices of Open Educational Resources (OER) to accelerate their study and adoption by subject matter experts.SPECTRUM+
Website LinkEarly in my time at UALR, I designed an AI agent named Athena to function as a synthetic “left brain for right-brained people.” Athena is a system prompt that can be delivered to an LLM such as ChatGPT, Claude, or Gemini to modify the default agent’s expertise and user interactions and ultimately function as an arts management assistant. This agent works well in conjunction or collaboration with trained arts educators, delivering personalized coaching based on a creator’s unique medium and stage of their artistic career. As knowledge work continues to be automated by AI at a rapid pace, I believe creative work will emerge as a new corridor of reliable, non-automatable labor, scaffolded in part by AI mentorship, management, and tools.Many creative people struggle to achieve full-time artistic success because mentorship and arts education are not equitably distributed in present society. This gap significantly affects the ability for artists to rely on their work commercially even if the quality and value of their work is high. This is doubly true for people of color, who have always been historically exploited by a predatory and highly entrenched system of arts marketing and distribution. The latter has resulted in artists of color disproportionately contributing to pop culture without equitable returns for their work. (“They love what we make, but not us.” - Anthony McCoy in DaCosta’s 2021 film Candyman)I shared this perspective with multiple arts educators and art institutions in the South, but I did not receive a response. So I formalized the idea of AI arts literacy as an instructional design and arts classification framework and published the core materials openly online as SPECTRUM+ while retaining ownership over the curriculum itself. I will continue to lead workshops on this topic in the near future and ensure that SPECTRUM+ extends to multiple languages and diverse artforms to support independent creatives who may otherwise be displaced by mass-generated corporate “AI art” or a dominant cultural lens.SULLIVAN + KELLER
GitHub LinkThe University of Arkansas at Little Rock does an incredible job of making learning possible for students of all physical and learning abilities. Accessibility is a core education principle for both my program and the school in which it is taught. One of the gifted students I met on campus has visual impairments that she developed late in life. When I reflected on the challenges that she and other members of the blind community face in everyday life, I realized that AI could act as a linguistic “sensory scaffold” if the user grants live-stream access to their smartphone camera.Sullivan and Keller are system prompts that work in the paid versions of the ChatGPT app to enable the default video-call assistant to function as assistive technology. Most accessibility apps are about communicating navigation or basic information about surroundings, and these are valuable pieces of technology that improve the lives of people around the world. (Visual impairment, like many divergent physical and learning abilities, is often a "disability" only in the sense that our shared physical and virtual environments have historically not been built with accessibility in mind even though we universally benefit from this design principle.)But it’s now possible to carry a video-enabled AI assistant around in our pockets, so it makes sense for us to produce smartphone applications that can communicate beauty as well.These AI agents are named after educator Anne Sullivan and her student Helen Keller, and they are respectively designed to A) interpret the immediate environment through a voice assistant’s lens of artistic creativity (User story: “If I could not see, I would want a creative person to follow me around throughout my day and narrate the space emotionally and aesthetically.”) and B) help a nonsighted person choose a wardrobe that expresses their personality and culture (User story: “If I could not see, I would want a fashion-savvy person to help me choose outfits from my wardrobe or suggest what to buy in a store in ways that express my style.”).The system prompts for these agents support multilingual communication and are split across two cognitive layers: one layer for communication in which the agent is assigned an area of expertise or “lens” through which to narrate what is in front of the user and a second layer of instructions that direct the agent how to visually interpret what appears on camera. The latter is the main innovation of this approach, as it scaffolds the AI’s ability to “think” about how sighted people interpret and interact with the environment to replicate a version of this cognition.I’ve shared Sullivan with social media communities for people with visual impairments online and received a great deal of constructive user feedback that I can use to refine the designs. For instance, I need to produce better documentation and onboarding instructions for people who use screen readers to interact with content on the internet. I hope to continue developing both of these prompts openly so that nonsighted people can modify the prompt scaffold and produce new tools, the need for which may not be immediately obvious to me as a sighted person.ALIENS
GitHub LinkMidway through my degree at UALR, I shared an AI agent with the Disability Resources Center (DRC). This agent is aware of the sentence-level challenges that are associated with reading for people with dyslexia. I’d realized that an LLM prompted with the specific context of paragraph- and word-level rules that make standard text “unfriendly” or intimidating for people with dyslexia can retain the core information and tone of a piece of writing (function) while transposing the sentences to maximize readability for this specific learner group (form).Please note that it may be possible for this kind of instructional modification to overscaffold or reduce the challenge of cognitive growth to the point of skill atrophy. It is important to constantly monitor a student’s progress on an individual basis under expert human supervision in accordance with best practices to ensure that learning is preserved.Once I’d tested the latter prompt, I realized that there is another ITS opportunity that leverages the same architecture.If a teacher can provide a single piece of writing or instructional materials for an entire class, an LLM can retain the core information and preserve key vocabulary while tailoring words and sentences to the exact reading comprehension needs of individual students.In essence, we can all read the same story but we will each receive it in a 1:1 format that is perfectly aligned with our target reading level. Many American classrooms (especially those in the American and Global South) are underfunded or understaffed, which makes it challenging for the teacher to provide individual feedback and materials appropriate across all reading levels. I believe instructors and their students would benefit greatly from curriculum delivered by this kind of Intelligent Tutoring System.I applied for an institutional grant to produce a beta version of the tool and studied the core concept of an adaptive ITS for reading instruction in some of my coursework (“ALICE”) but after I did not receive the grant, I decided to independently use AI to build a working version. I relied heavily on Claude 4.6 and ChatGPT 5.4 LLMs to write the code, as I’m not a software engineer.The resulting bilingual application (ALIENS - Adaptive Literacy Instruction for English and Spanish) can be extended to other languages to accelerate reading comprehension, so that everyone is reading in their continuously monitored zone of proximal development while still being able to be fairly assessed over the shared source material. It should be noted that the ALIEN reading system is not simply delivering the easiest form of a passage that a student can read; it is delivering the hardest words and sentences that a learner needs to get better without getting frustrated to the point of falling behind (ZPD).NUMERAIL
GitHub LinkOf all the professional and academic work I have completed up to this point, Numerail is the most important because it is a form of existential math. My research into Intelligent Tutoring Systems revealed that the current methods of deploying agentic AI in educational environments (and, more broadly, in professional and virtual environments) are highly unpredictable due to the underlying architecture of large language models (LLMs).LLMs produce text as the result of probabilistic sampling of millions of training materials, so there is always a nonzero chance that an LLM-powered AI agent will “hallucinate” (confidently produce a false or misinterpreted output) or be “misaligned” (take actions that violate policy issued via prompts or soft semantic guardrails). In order to safely deploy agentic AI, we need deterministic guarantees that a system cannot perform actions we’ve designated as harmful.I conducted extensive AI-assisted research into mathematical models of governance and found that convex geometry can be used as a gate for AI actions-as-numbers. Most if not all agentic actions are numeric parameters at the last mile (i.e. dollars passed along via a trading tool call or voltages sent to computer systems via an API), so we can define policies using geometry and place Numerail between AI models and the “buttons and levers” that affect the real world.This is like placing AI inside a “box” where it can still be creative or optimize while ensuring that only human-approved actions ever make it outside of that box. I’m not inventing the wheel here. This is an established design primitive present in fly-by-wire piloting systems, Simplex architecture, and NASA’s Runtime Assurance.As my educational background is not related to mathematics, I relied heavily on Claude 4.6 to write the convex geometry proofs using verification languages like Coq and Lean and build functional Python code on top of that foundation. I have shared Numerail with the UALR CS department, multiple AI safety organizations, and relevant faculty members at the top 10 academic AI research programs in the country.Numerail should be considered a work in progress; while I believe it shifts the hard problem of AI behavioral safety to the realm of engineering specification (a poorly written policy will be faithfully enforced by the system), it is necessary to consider all aspects of governance prior to deployment in production systems.I open-sourced the Numerail codebase under an MIT license because I believe AI safety belongs to everyone, and this field of study is often underresearched and underfunded within the current AI ecosystem.