ValidateNine

The problem

Real focus groups are slow and expensive, so most small teams skip them and guess. I wanted to find out how much useful signal you could get from a simulated version - and where that signal stops being trustworthy.

What I built

You submit a concept or a piece of messaging. ValidateNine generates a panel of distinct personas with different backgrounds and priorities, has each one react, and consolidates the responses into a structured report: what landed, what didn't, and where the panel disagreed.

Decisions that mattered

• Keeping personas genuinely distinct, so the panel doesn't collapse into one voice with different names.
• Structuring the output as a report a decision-maker can scan, instead of a wall of generated text.
• Being explicit in the product about what a simulated panel is and isn't.

I designed and built all of it: the product, the React/Vite front end, the Supabase database, the prompt architecture, and the deployment.