AI Education
I teach people, and the teams who train AI, how intelligence is actually built, one logic chain at a time. My classroom background turns complex AI concepts into skills learners can use right away.
Representative image only. The original interfaces, data and materials from this engagement are confidential and protected by non-disclosure agreements.
Project Snowman
Team leadership in AI evaluation
Applied a decade of classroom instruction to lead and mentor a team of AI evaluators, teaching the precise, evidence-based thinking that strong model evaluation demands. My approach treated evaluation as a learning discipline: one that can be modelled, scaffolded, and mastered.
Evaluating advanced AI is a skill that has to be taught. Evaluators must learn to think like the model they are testing, anticipate where its reasoning will break, and express their judgments in language precise enough for anyone else to reproduce. Talent alone does not guarantee those habits; they have to be built deliberately.
Teaching taught me that every wrong answer has a cause. A student’s misconception and a model’s hallucination follow strikingly similar paths: a missing link, a faulty assumption, or a rule applied in the wrong context. The goal of this work was to bring that diagnostic lens to the team, so evaluators could understand not just that a model failed, but why.
Backward Design for Evaluation Standards
In teaching, strong lessons begin with the end goal and work backward. I applied the same principle to evaluation: starting from what reliable training data must look like, then identifying the skills evaluators needed to produce it.
The Prompt Playbook became the team’s curriculum, turning complex requirements and recurring edge cases into clear explanations, worked examples, and models of strong practice.
Diagnosing Reasoning Like a Teacher
I taught evaluators to read a model’s answer the way I once read a student’s essay: following the chain of logic step by step and pinpointing the exact moment it went wrong. Was the failure caused by missing knowledge, a misread instruction, or a reasoning leap the evidence could not support?
Treating model errors as misconceptions to be diagnosed, rather than mistakes to be counted, helped the team write sharper prompts and fairer, more defensible rubrics.
Feedback That Builds Independent Judgment
My reviews followed the principles of formative assessment. Rather than simply correcting work, I explained the source of each problem, showed how to fix it, and asked evaluators to apply the reasoning themselves in their next task.
I tracked each writer’s growth over time and gradually released responsibility as their judgment strengthened, the same scaffolding I once used to move students from guided practice to confident, independent work.
Teaching How AI Thinks
Some of the most valuable lessons were about cognition itself: how logic chains are constructed, why a confident answer can rest on a false premise, and how small ambiguities in a prompt can send a model down the wrong path.
Helping evaluators think about how a model thinks, a form of metacognition, turned rule-following into genuine understanding and made the team more adaptable when new edge cases appeared.
The team developed stronger, more independent judgment in prompt engineering, rubric design, and failure analysis, producing evaluations that were more consistent and easier to defend. The Prompt Playbook remained a shared reference that new and experienced evaluators alike could return to.
The work demonstrated how the principles of good teaching translate directly to AI: when people understand how intelligence is built, they become far better at testing, training, and improving it.
Representative image only. The original interfaces, data and materials from this engagement are confidential and protected by non-disclosure agreements.
Project AI Classroom
Private virtual classes
Teach private virtual classes in AI literacy, prompt engineering, and problem solving, drawing on more than a decade of English teaching and hands-on experience training advanced AI models.
Many learners now meet AI as a shortcut: a way to finish homework or tasks without the thinking those tasks were meant to build. Used that way, AI becomes a crutch. Used well, it becomes an amplifier for curiosity, clarity, and independent thought.
These classes teach learners to use AI as that amplifier. The aim is not simply to produce better outputs, but to develop the critical thinking, precise communication, and problem-solving habits that make someone genuinely capable with, and without, AI.
Diagnostic Starting Point
Every learner begins with a short diagnostic of how they currently read, write, reason, and use AI tools. Lessons are then built around their goals, whether that is stronger schoolwork, a professional skill, or confidence with new technology.
Prompting as a Writing Skill
Prompt engineering is taught as an extension of clear writing: defining purpose and audience, setting constraints, supplying context, and revising based on results. Learners see first-hand how a vague request produces a vague answer and how precision changes everything.
Thinking With AI, Not Through It
Learners practise questioning AI rather than accepting it: checking claims against sources, comparing answers, spotting flawed reasoning, and recognizing hallucinations. These are the same habits I use to evaluate advanced models, adapted for everyday use.
Applied Problem-Solving Projects
Each unit ends with a practical project in which learners use AI to research, plan, and solve a real problem, then explain their reasoning in their own words. The final product always reflects the learner’s thinking, with AI as a tool rather than the author.
Learners leave with practical AI skills alongside stronger reading, writing, and reasoning. They become more discerning users of technology: able to get better results from AI, and better equipped to recognize when its answers fall short.