Library Scout

It takes multiple clicks and scrolling just to find the events page. Once there, users are dropped into nearly 8,000 records with minimal guidance. The UI asks people to go on a treasure hunt—without a map.

By the third click and first scroll, most users aren't looking for a class anymore. They're looking for the exit.

The Cultural Moment

I saw this as an opportunity. Also, as of early 2026, we're seeing a cultural shift toward “building your own curriculum,” analog hobbies, becoming—disgustingly educated, and third spaces surging among Gen Z.

It's a response to automation fatigue and AI-generated content overload. Libraries already support this movement perfectly.

Building a New Taxonomy

The library's existing categories weren't semantically aligned with how people actually browse or think about learning. So I built my own taxonomy—one rooted in identity and intent, not institutional structure.

I introduced timely, human categories like Cozy Crafts, Learners, and DJs & Producers, then broke them down into Archetypes, Categories, and Subcategories. I added a touch of academic familiarity—labels like Art 101—to make exploration feel approachable rather than overwhelming.

I scraped the entire February class catalog, used AI to assist with tagging, and then manually reviewed and corrected inaccurate records. To make this repeatable, I built a custom classification prompt directly from the taxonomy I designed.

BOOM — I had built my own cheat sheet.

I felt like I was back in college: the person who makes the study guide and sends it around. Except this time, even when I handed people a clean spreadsheet, they still couldn't find what they needed.

Because the truth is—not everyone is good at spreadsheets. The categories helped (and they worked well for targeting the right people on TikTok), but the format itself was still friction-heavy.

...and by stepping away, I was able to understand my goals and high-level features with much more clarity.

If AI was going to build this correctly, I needed to define the DNA of the system before touching a single line of code.

Defining the System with OOUX

One of my favorite tools for this is Object-Oriented UX (OOUX)—a practice that models digital systems around the objects users actually interact with. It breaks complex technical systems down through a design lens, much like how the DOM structures interfaces, making backend modeling more intuitive and aligned with real-world use.

OOUX starts with identifying the core objects in the system—the “things” users come to interact with. In this case, those objects were:

The diagram above shows an object block. Blue represents the object itself, pink indicates attributes (data), yellow represents qualitative information, and light blue shows nested objects.

CTA inventories are especially powerful here. They expose gaps in requirements, surface unclear assumptions, and force alignment early—helping teams ask better questions long before implementation begins.

From this view, I could already see the logic, states, and pages I would need to build. I also knew this would be valuable context to feed into the AI.

To deepen that context, I created a user flow map to understand how users in different states would move through the system.

Lastly I made a simple site map so the AI was crisp on how things connect.

Refining the Data Model

Around four hours in, I got a little impatient and shifted focus to improving how I was collecting and modeling the data. I started questioning whether my categories were actually the best ones —and whether there was a better way to structure them.

I spent time working with Gemini to analyze the data and anticipate user needs, refining the categories so they were broad enough to scale but not so broad that they obscured what was actually being offered.

To support more accurate tagging, I also created an AI-readable column of keywords designed to guide the model toward the correct classification decisions.

This part was humbling—and a reminder of why data science is its own discipline.

I learned quickly that tagging performs better in smaller chunksand benefits from scripts and confidence scores to reduce manual review. While I want to reduce my hands-on involvement in this step long-term, it's an area where I know I'll need additional support or automation.

Library Scout went live in February 2026 with no launch campaign, no paid promotion, and no marketing budget.

The most telling signal was the referral traffic breakdown. Google was the top source. That's organic search finding the app without any SEO work, which means the problem has real demand that exists independently of social media reach.

I've been building Library Scout in public, sharing the process and testing openly on TikTok. That approach has given me something more useful than a traditional feedback cycle: a running signal on what resonates, what confuses, and what people actually want to do with it.