Reflection
What This Project Taught Me
A first-person look at the design decisions, tensions, and lessons behind building AI that serves learning.
The hardest decision was restraint
The temptation with AI is to let it do everything — answer questions, grade essays, drive the lesson. The instructional-design discipline was the opposite: deciding where AI genuinely helps and deliberately holding it back everywhere else. A coach that gives answers feels helpful and produces nothing; a coach that scaffolds feels slower and produces thinkers.
Personalization is a design problem, not a tech problem
The five paths, the recommendation logic, and the adaptive routing matter far more than the model behind them. Good personalization is mostly clear learning progressions, strong assessment, and humane defaults — AI just makes them responsive in real time.
Teachers must stay in command
Every recommendation is a suggestion with a visible rationale and an override. The moment AI changes a child's path without a human's sign-off, you've traded professional judgment for automation. Designing the override workflow was as important as designing the lessons.
Trust is built with transparency
Families and educators trust what they can see. Disclosing when AI is used, logging the coach, and auditing for bias aren't compliance chores — they're the foundation that makes adoption possible.
What I'd improve next
- Integrate a live, safety-filtered LLM with full teacher-visible logging.
- Run a classroom pilot and study where the coach helps vs. frustrates.
- Add reading-fluency analysis and bias-audited early-warning models.
- Co-design the path movement rules with practicing teachers.
What it demonstrates
This project shows I can integrate AI responsibly into K–12 learning, design personalized digital experiences, apply analytics to instruction, and lead the kind of digital transformation a district or ministry can actually trust and adopt.