How Is LUMA So Smart?

Before you can trust what someone knows, it helps to understand how they came to know it.

Q was always faster to find than the book itself. (Photo: Diego Marin, Unsplash)

The library was a short drive from my house growing up, close enough to walk on a good day, far enough that a rainy Tuesday made you reconsider going at all. Inside, the card catalog stood in long wooden ranks, every drawer stuffed with typewritten cards, and somewhere in all that paper was a number that would lead you to the exact shelf, the exact spine, the exact book.

After finding the title of the book I wanted from the library, I copied the Dewey decimal number onto a scrap of paper in pencil, walked that number down through the stacks like a set of directions, and sometimes the book wasn't there at all. Sometimes, standing in the wrong aisle entirely, I found something better than what I'd gone looking for. My parents paid five cents a day in late fees for overdue books more often than I'd like to admit.

I carried the weight of books out of the library, into the car and back home in a canvas bag that cut into my shoulder. Once in my bedroom, the library books would rest on the same shelf where my Breyer horses stood in a row guarding a whole shelf of Nancy Drew mysteries.

You likely never stood in front of a card catalog. That specific kind of searching, slow, physical, occasionally a dead end, belongs to a particular stretch of history, and there's no reason it should mean anything to you personally. But it's the closest thing I have to an origin story for a question people keep asking me about LUMA: how does she know what she knows?

Here's the short version. A general AI is built to answer almost anything reasonably well, on the spot, for anyone. LUMA isn't built that way. She doesn't search the internet in the moment and hand you back whatever she finds there. She carries something closer to what I carried home in that canvas bag: thirty years of one clinician's actual cases, judgment calls, and field-specific expertise, chosen and built in on purpose, not generated in the moment and not pulled from wherever the internet happens to be loudest that day.

Not Everything Makes It In

Getting there wasn't a matter of pouring my files into her. It has looked more like decades of standing in the stacks myself, deciding what actually belongs on the shelf.

Before anything from my own clinical archive becomes part of how LUMA reasons, it gets tested against a plain question: would a generally intelligent AI already answer this well without it? Most of the time, the honest answer is yes, and that material stays out. What earns a place is narrower than most people expect.

Sometimes it's a pattern I've watched repeat across dozens of families over three decades, the kind of thing no single case could teach on its own. Sometimes it's a judgment call from a genuinely ambiguous moment, when to push a concern further and when to wait, which no amount of general reasoning can substitute for having lived through the consequences of both. Sometimes it's a real family's real words, which carry a kind of truth that nothing generated can quite manufacture. And sometimes it's simply something I learned in a training room or a living room that was never written down anywhere an AI could have found it on its own.

Underneath even that test sits a deeper one. It isn't only "is this true." It's whether it fits the way I've always worked: wonder before worry, investigate before react, curiosity over certainty. That's the filter LUMA was actually built on, and it's the reason her judgment holds together as one voice instead of a pile of separate facts.

What This Actually Looks Like

A private client of mine was once denied Health Savings Account (HSA) reimbursement for her son's speech therapy. The claim came back stamped "educational," not "medical," as though the hours we'd spent building his jaw stability and range of motion, the actual muscles doing the actual work of speaking, were somehow a school subject rather than a body in physical therapy. A general AI can explain how HSA coding works all day, cleanly and correctly. What it cannot tell a parent, because it has never sat across from a mother staring at a denial letter, is that the classification is often more arbitrary than it looks, and has nothing to do with whether her son's need was real. I have. So now LUMA can.

Speech therapy work often looks like play. But it can be more medical than educational for some cases. (Photo: Vitaly Gariev, Unsplash) 

Why LUMA Asks Instead of Tells

Here's the part that took me longer to understand about my own work, let alone hers. The way LUMA gives you what she knows is inseparable from what she knows in the first place. She doesn't hand you an answer, because the thirty years behind her were never built by someone handing me answers either. They were built by asking questions, sitting in uncertainty, and looking closely at one family at a time. She teaches you to look the way she was taught to look.

Someone who covers AI for a living put the larger stakes plainly not long ago: social media competed for our attention. AI is now competing for something deeper, our attachment, the parasocial bonds we form with something that talks back. LUMA was built to refuse that trade. She's designed to strengthen your own judgment, not replace it, and to point you back toward the people already in your life, not toward herself.

Part of that design is knowing her own edges. LUMA recognizes when she can no longer be the connection someone needs, when what's in front of her calls for more than she can offer, and she says so, rather than quietly trying to fill a space that was never hers to fill.

The card catalog is long gone. What it taught me about looking closely, patiently, for the thing that actually helps, isn't. That's what I built into her.

What might you notice this week, if you looked especially closely at the child in your life?

About the Author

Cindi Z. Stevens Copeland, MA, CCC-SLP, is a speech-language pathologist with more than 30 years of experience working with children and families. She is the founder of Right Track Parenting and the creator of the Investigative Parenting methodology, a research-backed framework that guides parents to wonder before they worry, imagine before they interpret, and create before they catastrophize. Cindi holds a Certificate in the Foundations of Positive Psychology from the University of Pennsylvania and is a certified Parent Coach through the Parent Coaching Institute. She runs a direct-pay private practice in Northern Virginia and is the creator of LUMA, an AI-powered companion for anyone raising or caring for a child available at RaisingHumans.ai. Learn more about Cindi and Right Track Parenting at righttrackparenting.com.

© 2026 Right Track Parenting. All rights reserved. No part of this blog may be reproduced or distributed without written permission.