Persona Atlas: Mapping How Famous Minds Think

TL;DR — Persona Atlas turns a public figure into a behavioral portrait you can actually measure. Type a name, and a small-model agent goes off to research that person on the open web, assembles a grounded dossier, and then answers a fixed set of open-ended "thinking" questions in their voice. Every answer is embedded, so a persona stops being a wall of prose and becomes a point in space — and suddenly you can line several thinkers up shoulder to shoulder and see whose mind reaches for skepticism, whose for humor, whose for cold abstraction. The wager underneath it all: personality is a matter of style, not horsepower — so it survives even when the models doing the talking are small. Which, this being the build-small hackathon, is rather the point.

▶️ Watch the short tour — research a persona, compare a few, read the trait heatmap.

▶️ Watch the short tour — research a persona, compare a few, read the trait heatmap.

What if you could put Socrates, Churchill, and a Silicon Valley founder in the same room, hand them the same unanswerable question, and just watch how differently each one reaches for an answer? Most benchmarks measure what a model knows. Persona Atlas is curious about something slipperier — how does a given mind move? — and it tries to make that visible instead of merely asserting it.

A run is three steps.

First, research. A tool-calling agent issues real web searches, pulls a portrait, and assembles a public profile, a list of grounded facts — each one linked back to a source it actually visited — and a "style hypothesis": its best guess at how this person attacks a problem they've never seen before.

Second, the persona answers the benchmark — ten deliberately open-ended prompts about identity, ethics, truth, free will, meaning, and machine consciousness. There are no right answers on purpose. These are the questions where a personality leaks through, instead of the model's raw capability.

Third, every answer becomes an embedding. Once a persona is geometry, you can do something prose never lets you do: put two of them next to each other and measure the distance.

Pick any of the saved personas and the comparison view does two things. It measures how far apart their answers sit in embedding space — one number for how much the whole group diverges — and it scores each persona against ten trait anchors (meticulousness, clarity, creativity, skepticism, confidence, kindness, humor, curiosity, pragmatism, abstraction), painted as a trait-leaning heatmap.

The grid is double-centered, which matters more than it sounds. A warm cell never means "high on this trait" in some absolute sense — it means this persona leans toward that trait more than the others you happened to put on the table. Drop a handful of very different people side by side and the rows pull apart: one running warm on humor and confidence, another on abstraction and skepticism, often in a way you can feel a beat before you can put it into words. That gap — between sensing a difference and naming it — is exactly what the map is for.

An earlier version had math problems and trivia, with right answers and a leaderboard. All of it got cut. A correct integral looks the same whether the persona is Einstein or anyone else — objective tasks measure the model, not the person. What's left is only the material where stance, tone, and reasoning style actually diverge. The output is a stylistic mirror, not psychometrics: it shows what a persona's answers resemble, relative to the others — never a reading of the real human's soul.

Everything runs on small, hosted models through Hugging Face Inference Providers — a compact generator driving the agent, a lightweight embedding model doing the geometry — plus live web and image search for grounding. The front end is Gradio, with three tabs: research a run, compare saved personas, and inspect the full agent trace, so you can check for yourself that it's leaning on real sources rather than quietly making things up. A set of personas ships prebuilt, so the comparison is explorable the moment the page loads — no token required.

A portrait is only as good as what the open web surfaces, and name collisions are real noise. The map captures a persona's answering style under one model on one day — not a person, and certainly not a verdict. Read it the way you'd read a caricature: it exaggerates to reveal, and it's most useful as a lens for contrast and conversation.

Open the Compare saved personas tab to start, or research someone new and add them to the atlas: huggingface.co/spaces/build-small-hackathon/persona-atlas