Human Behaviour Simulation

What if you could test a message on a million people — before a single real person ever saw it?

That's this project: a human-simulation layer that models how real audience segments think, feel, and react. Not guesswork, not vibes — every simulated audience is built from real survey data about the people it represents, so the "digital crowd" behaves the way the real crowd actually would.

It works in both directions:

• Test a message. Hand it something you're about to say, and it predicts how each segment will react — who nods along, who pushes back, who quietly changes their mind.
• Generate a message. Hand it a goal instead, and it writes messaging calibrated to move a specific segment toward that goal.

It goes deeper than reactions. The system measures each population group's firmness, susceptibility, and persuadability — how likely their views are to shift, not just in response to a message, but under broader societal, political, economic, and cultural currents. Some audiences are rock; some are sand. Knowing which is which changes everything.

The honesty layer is the part I'm proudest of. Every prediction ships with error bars, is checked against a placebo baseline, and passes through a reliability gate before it's allowed to count. The system says plainly what it can and cannot know — separating robust predictions from speculation instead of dressing both in the same confident voice.

It pairs beautifully with the Council of Experts. CoE ends where this begins: it distills a mountain of data into proven, fully-cited findings — and those findings are exactly the kind of ground truth these simulated audiences are built on. One system proves what's true today; this one predicts what people will do next.

Under the hood (for the engineers): LLM-driven population simulation over structured survey representations, digital audience modeling, uncertainty quantification, placebo-controlled evaluation, and reliability gating.

Why it matters (for everyone else): it replaces "let's launch it and hope" with "let's simulate it and know — error bars included."

Skills: Numerical Simulation · Large Language Models (LLM) · Uncertainty Quantification