The math behind the mirror.
How computational analysis turns conversations into cognitive portraits.
Lumensis uses methods from natural language processing and information theory to find patterns in your AI conversations — patterns that are invisible from inside them. Here's what we measure, and how.
Five dimensions of thinking.
Attention Distribution
What pulls your focus — when nobody's directing it?
We use BERTopic clustering to map where your cognitive energy goes across conversations. The result: a topographic map of what you actually care about, not what you say you care about.
Learning Rate
How fast does your thinking evolve?
We track semantic displacement — how your conversation topics shift over time. High velocity means rapid exploration. Low velocity means deep commitment to specific territory. Neither is better.
Loss Landscape
Are you stuck, or transforming?
Inspired by optimization theory: we measure topic diversity against transition predictability. The result reveals whether you're in a plateau, a valley, or approaching a phase transition in your thinking.
Coherence
How integrated is your mind?
Cross-domain embedding similarity shows whether your different interests inform each other or exist in separate silos. High coherence means your thinking has internal structure. Low coherence means you're exploring in parallel tracks.
Gradient Signal
Where is your thinking actually heading?
Directional analysis of topic evolution over time. Are your interests converging toward something specific, or diffusing outward? This is the trajectory beneath the noise.
What this is — and isn't.
Lumensis is conversational pattern exploration, not a validated psychological assessment.
Our methods — BERTopic topic modeling, KL and Jensen-Shannon divergence, transformer embeddings, entropy analysis — are mathematically rigorous. They measure real patterns in real data.
What they don't do is map neatly onto established psychological constructs. We haven't run external validation studies (yet). The metrics describe your conversational behavior, not your personality.
The real insight comes from what these patterns reveal when you see them together — and from the AI-powered narrative layer that translates numbers into observations about your thinking.
Think of it as a new kind of mirror: precise about what it shows you, honest about what it can't.
Have questions about our methodology? hello@lumens.is