Persona Validator Tool - Most personas are invented. I built one that proves itself.

Live tool: https://personavalidator.app

Walk-through trailer:

https://www.youtube.com/watch?v=UXk6wmmDh38

The problem I kept running into

Personas are supposed to anchor product decisions. In practice, most are invented. Someone writes up a plausible-sounding user, gives them a name and a stock photo, and the team treats that fiction as evidence for months. When a stakeholder asks "how do we know this persona is real," the honest answer is usually that we do not.

That gap bothered me enough to build a tool that closes it. Persona Validator turns real user feedback into personas where every claim traces back to the exact responses that produced it, and where the predictions are tested against the data instead of confirmed by wishful reading.

The idea: treat persona work like a hypothesis, not a story

The core of the tool is a validation arc borrowed from how researchers guard against fooling themselves.

Before any analysis runs, you write your persona hypotheses and lock them. The lock is permanent and timestamped, so the bets provably predate the evidence. This is preregistration, and it is the difference between rigor and the look of it. Anyone can produce personas that match what they expected when they get to write the expectations after seeing the data. Locking first removes that temptation entirely.

Then the tool does the analysis, and the analysis alone decides what is true. It reads the responses, codes each one into behavioral signals, clusters them by behavior, and synthesizes personas from the patterns. Finally it compares what you predicted against what the data actually produced, and labels each prediction: confirmed, revised, rejected, or emergent, meaning a persona the data revealed that you never predicted.

A rejected hypothesis is not a failure of the tool. It is the tool working.

The discipline that holds it together

A few rules keep the output honest, and they were the hardest and most interesting part to get right.

Nothing gets invented. If the responses contain no demographic signal, the persona stays behavioral only and says so. The tool refuses to fabricate an age or a job title to make a persona feel complete.

Every attribute is labeled by how it was grounded: taken directly from the text, synthesized across several responses, or inferred as a flagged reach beyond what was said. Inference is allowed, but never disguised as fact.

Every claim links to its supporting responses through a traceability matrix. Pick any persona attribute and you can follow it back to the exact reviews behind it, and see whether it came from primary data (a survey) or secondary data (public reviews). A claim supported by both is stronger, and the tool surfaces that.

The name and photo on each persona are labeled as illustrative, never derived from data. They humanize the card without pretending to be findings.

The flagship study: Jira

To test the tool on something I understand from the inside, I ran it on Jira, the project tracking tool many business analysts live in. It is heavily reviewed and famously polarizing, which makes for real clusters with real friction rather than bland praise.

Before looking at anything, I locked five hypotheses about who reviews Jira and why, drawn from years of working in and around it:

• The Reluctant Developer, who uses it because their team mandates it and resents the friction.

• The Overwhelmed Admin, who configures it for everyone else and struggles with the complexity.

• The Mobile-Frustrated User, whose complaints are about the phone app specifically.

• The Small-Team Escapee, who finds it crushingly over-built and is on the way out.

• The Migration Casualty, whose review is driven by a disruptive pricing or platform change.

Then I ran the pipeline and let the data decide.

What this demonstrates

I built this as a portfolio piece, so it is fair to ask what it is meant to show. It is less about the code and more about how I approach ambiguous problems:

I start from a problem worth solving, not a feature I want to build. The problem here was that persona work is rarely defensible.

I care about evidence and traceability, because requirements that cannot be traced to a source are opinions.

The traceability matrix is the same discipline applied to research.

I design for intellectual honesty over a flattering result. The preregistration lock and the grounding labels exist specifically to stop me from fooling myself or a stakeholder.

Lastly, I can take an idea from problem to a deployed, working product, including the requirements artifacts a team would actually use: a business requirements document, user stories with acceptance criteria, and exportable traceability, all generated from the study.

Limitations, honestly

Public app store reviews skew toward grievance, so they over-represent unhappy users. The tool addresses this by supporting primary survey data alongside reviews and flagging which personas appear in both, but a study built on reviews alone should be read with that skew in mind. The personas are behavioral, not demographic, because the source data rarely contains reliable demographic signal, which is a deliberate limit rather than a gap I overlooked.

Try it

The tool is live at https://personavalidator.app, opening on the Jira study so you can trace any claim back to its evidence yourself. The walk-through is at https://www.youtube.com/watch?v=UXk6wmmDh38.

It is built with Next.js, MySQL, and the Anthropic API, and deployed on a Linux server behind Nginx. I am happy to talk through any part of the design.

Contact me:

Email: chadwambles@gmail.com

Linkedin: www.linkedin.com/in/chad-wambles-071918191