XrÆ

The problem nobody is solving

Every day, millions of people consume media that has been carefully crafted to shape what they think. Not through lies — through technique. Word choices that trigger emotions. Framing that predetermines conclusions. Selective context that changes the entire story. These patterns are invisible unless you know exactly what to look for.

Generic AI can't do this. You can't prompt your way to reliable influence detection. General-purpose language models hallucinate detections, miss subtle patterns, and lack the consistency to analyze thousands of articles against the same standard. The problem demands a specialized machine.

So we built one

XrÆ is a 70+ billion parameter large language model, fine-tuned from the ground up for one job: detecting persuasion techniques in text. It doesn't summarize. It doesn't chat. It reads media and identifies exactly how that media is trying to influence you.

Not a prompt. Not an API call. A real model.

Most “AI-powered” tools send your content to a generic model with a clever prompt and hope for the best. That approach is brittle, inconsistent, and fundamentally limited by what a prompt can express.

XrÆ is different. It doesn't follow instructions about influence detection — it understands influence detection. The knowledge isn't in a prompt that can be copied. It's embedded in the model itself, learned from thousands of real-world examples until the patterns became instinct.

The result is something that thinks like a trained media analyst, works at machine speed, and never gets tired, distracted, or influenced by its own biases. The same rigor, applied to every piece of content, every time.

Analyzed once is not analyzed

Most tools analyze your content once and move on. We think that's lazy. Every piece of media that XrÆ touches goes through eight independent analysis passes. Eight times the model reads the same content, looking for what it missed the last time.

Why? Because influence techniques are designed to be subtle. The first pass catches the obvious. The second catches the structural. By the eighth, the things that were engineered to slip past you don't slip past us.

Nobody does this. It's computationally expensive and most people don't care enough to try. We do — because if you're going to trust an analysis, it should be one that was obsessed over, not one that was guessed at.

A machine that gets smarter every week

Persuasion techniques aren't static. New narratives emerge. Framing strategies adapt. What worked last year doesn't capture what's happening today.

XrÆ runs a compound learning cycle: new media is transcribed, analyzed by expert-level AI, and fed back into training data. Each week, the model retrains on everything it has ever learned plus the latest patterns. Detection accuracy doesn't plateau — it compounds.

Curriculum training targets the specific patterns the model struggles with, drilling on hard examples until they're mastered. The techniques the model missed last month are the ones it catches this month.

Zero cloud dependency

XrÆ runs on dedicated GPU hardware. No cloud API. No third-party inference service. No rate limits. No data leaving our infrastructure.

This isn't a cost optimization — it's a design choice. Full control over the model, the data, and the pipeline means we can iterate faster, serve analysis at scale, and guarantee consistency across every piece of content we analyze.

What XrÆ powers