I have big ideas. Urgent ones at 11 p.m., usually gone by breakfast. So when the thought hit me — an AI reading recommender for our school library to actually get kids excited — I asked my partner if I could trade my weekends for it.
“Just go for it,” she said
May 2025 conversation. November app live in class. Fast track? No. Painful slog. Yes.
The Why
I’ve taught in the U.K., the Middle East, Southeast Asia. Eleven years. I’ve seen librarians curate perfect books that sit on shelves, gathering dust. Why? No system connects this child to that book.
Existing tools are expensive. Rigid. They ignore what’s actually on your shelf. I realized the tech existed, but nobody built it for a teacher like me. So I did. I used vibe coding.
Learning the Hard Way
Vibe coding. You describe the code in plain English to an AI. You don’t code. You talk. I started talking.
Progress was glacial. Day forward, three back. I almost quit all summer. I was running on a twelve-year-old Mac. Just installing the environment felt like a second job. Then I lost hours of code. No backup. Just stared at the screen. Empty.
Book covers were the nightmare. I had ten thousand titles. I needed legal APIs, no scraping. Wrote code until I broke. Got it to work, badly. Built a manual check page. Weeks of clicking. Then that page crashed. Restarted from zero.
Switched from Copilot to Claude. Better, but still crazy. Loops. Errors. Still better than nothing.
Now? What took me months takes hours. LLMs move frighteningly fast.
What takes me days and weeks in late 5, I can accomplish in hours. The rate is frightening.
How It Runs
Keep it simple.
1. Upload school catalog as CSV. Done. No re-cataloging.
2. Make student profiles.
3. Run a quick reading check. Level, interests.
AI scans the catalog. Matches it to the kid’s likes, favorite authors, class topics. Outputs a list of books already in your library.
Profiles show name, reading age, genre prefs, current topic. Data stays with the teacher and librarian. Privacy matters. LibraryAid is GDPR and COPPA compliant. Google Firebase storage. No email addresses. Login is code + PIN.
Kids log in. See fifty recommendations. Ranked.
* Mark “reading”
* Mark “finished”
* Mark “want”
Finish a book? Write a review. Answer comprehension questions. Get points. Unlock worm accessories. Different genre points for different worm gear. Reading wide beats reading volume here. Reviews feed the engine. One kid’s discovery becomes visible to the school.
The engine uses a “master books” list of 1,000+ award titles. It doesn’t just match level. It digs for the hidden stuff kids never find on their own. Sometimes it’s a new genre under a familiar topic. Sometimes it’s the next logical step. A similar author. A series continuation.
People Actually Liked It?
My colleague said I could do it. Honestly. That confidence meant more than any tutorial. Other teachers gave feedback. Frank. Humbling. One integrated it immediately. Called it useful.
My son loved it. Tested it relentlessly. Told his school. Then did what only a twelve-year-old could do. Asked an LLM if LibraryAid would succeed. It said “yes.” Contemporary parenting moments.
The Kids Reacted
Something shifted. Unenthused readers wanted the library. It became a treasure hunt.
One English learner, two grades below his placement. Matched him right. He read three times the average progress of his classmates. Tech didn’t fix him. It just gave him books worth the struggle.
I read aloud Swimming Against the Storm. Environmental theme. Striking impact. Next day? The class swarmed the app for adventure books like that. Human spark. Machine fuel. The app works best beside a teacher. Not instead of one.
What Debugging Taught Me
Fixing code and fixing misunderstanding look alike. Both need patience. Both need hypotheses. Writing adaptive algorithms forced me to think about differentiation.
Most edtech fails. Built for admins, not teachers. Optimizes for dashboards. Ignores thirty distinct children.
The tool surfaces books. It does not guarantee inspiration. One kid showed me her list. Eyes pleading. Lost. The list was great. Classics, gems, comfort, stretch. She only saw the series she knew. The algorithm worked perfectly. She needed a person.
You cannot code trust.
There isn’t a recommendation engine that replaces the moment a child asks, “What do I read now?” and looks at you.
Build tools that extend us. Let AI match. Let us nudge. LibraryAid might be the best thing I’ve ever made. Even better than some lessons.
But it’s not a magic wand.
What do we do when the screen says “try this” and the kid looks blankly at us?
That part isn’t automated yet. Maybe it never should be. 🐛
