A junior year project built with AI. Here's what I designed, what AI built, and what I learned.

The Problem That Started Everything

In shared minds class writing assignment, we were asked: what's broken about the platforms we use every day?

My answer was simple. Algorithms decide what we see. On Instagram, Xiaohongshu, and TikTok, we scroll through an endless feed curated by engagement metrics -- likes, comments, saves, shares. We consume whatever the algorithm serves us, like animals being fed, often spending hours without realizing it. And creators judge their own work by these same metrics, chasing numbers instead of meaning.

But the deeper problem is what we miss. Sometimes I want to find someone who shares my exact perspective but arrived there through completely different reasoning. Or someone who disagrees with me entirely but argues their case so well that I have to rethink my own position. These people exist. Their posts exist. But the algorithm buries them because engagement metrics can't capture intellectual resonance.

So I asked myself: what if we connected people by the meaning of what they say, instead of how many people clicked a button under it?

The Design Idea: A Semantic Star Map

I imagined a dark canvas -- like a night sky -- where every post is a star. Your position isn't determined by popularity. It's determined by what you said and how you said it.

I defined two axes as a starting point:

• X-axis: Radical vs. Rational. Not about political stance, but about expression style. Aggressive language, profanity, excessive punctuation, pure emotional venting -- that's the radical end. Structured arguments with cause-and-effect reasoning -- that's the rational end. A person can hold a moderate opinion and express it aggressively, or hold an extreme opinion and express it calmly. This axis captures how you communicate, not what you believe.
• Y-axis: Positive vs. Negative. This one is more straightforward -- the emotional valence of the text.

I'll be the first to admit: two dimensions aren't enough to represent someone's "opinion position." My real vision is bigger -- within a specific debate topic, map where each person's viewpoint falls relative to others arguing about the same thing. That's a controlled variable. Throwing random posts about food, politics, and cat videos onto the same 2D plane is not meaningful. But mapping responses to "Should AI replace teachers?" onto a shared semantic space -- that could actually reveal the structure of a conversation.

And here's where it gets interesting: topics could connect to each other. A post sitting at the edge of one topic's star map might overlap with another topic entirely. Imagine dragging across the map like navigating a globe. As you move past the boundary of one debate, the title shifts, and you've drifted into an adjacent conversation. The map becomes alive -- not a static grid, but a landscape of interconnected ideas.

I haven't figured out how to implement this yet. I'd need to define the right parameters first. But the concept feels right.

The Engineering Constraint: Why Not Just Use an LLM?

The obvious approach would be: send every post to GPT or Claude, ask it to score the text on multiple dimensions, and use those scores as coordinates.