AI UX Research Toolkit GitHub ↗

Portfolio

Augmented Rigor
& The Co-Researcher Model

UX research tools that treat AI as a tireless co-researcher—not a shortcut. Each tool targets a specific moment where human judgment is most at risk.

🧠 My Philosophy

To enhance the depth and objectivity of my work, I treat Artificial Intelligence as a tireless "co-researcher" rather than a simple automation tool. My goal is not to work faster, but to leverage AI to challenge my own assumptions and ensure the highest quality of insights.

The tools below are concrete proof of that philosophy—each one designed to introduce friction at the right moment, not remove it. They exist because the hardest part of research isn't collecting data; it's knowing when you're wrong. One tool is live today; the rest are in development—some as working Streamlit apps, others as documented concepts with draft co-researcher prompts on GitHub.

Tools

One live app you can run today. Three more in development—each folder documents the UX problem; some include a working Streamlit app, others a draft co-researcher prompt.

Live tools

01

The UX Research Interview Refiner

Live

Before the interviews begin

The UX Problem

Interview guides often carry hidden assumptions, leading questions, and coverage gaps that only surface after sessions are underway. By then, revision is costly—and bias has already shaped what you hear.

How It Helps

A Streamlit co-researcher takes your objectives and draft guide, generates an independent parallel guide, then reports on blind spots, framing bias, and missed angles. Supports Claude (recommended) or Gemini. You provide your own API key — cost and terms depend on your provider account, not the app host.

Try live demo View on GitHub Report an issue

Not air-gapped: analysis sends your draft to Anthropic or Google. For confidential work, run locally. Hosted demo also passes data through Streamlit's servers. Hosted demo may take ~1 min to wake if idle — click through the sleep screen.

In development

02

Thematic Analyzer

In development

During synthesis · Streamlit app in progress

The UX Problem

When a single researcher codes interview transcripts, confirmation bias is inevitable. We notice what we expect to find and quietly deprioritize contradictions—the very signals that often contain the most valuable insights.

How It Helps

A Streamlit co-researcher runs a blind thematic analysis on your transcript, then compares it to your working themes and reports only divergences: missed themes, underweighted quotes, and framing differences. Supports Claude (recommended) or Gemini. You provide your own API key — cost and terms depend on your provider account, not the app host.

View on GitHub Report an issue

Run locally with .\run.ps1 — hosted demo not deployed yet. Not air-gapped: analysis sends your transcript to Anthropic or Google.

03

Research Plan Challenger

In development

Before the fieldwork begins · prompt scaffold

The UX Problem

Research plans often carry hidden assumptions—about user behavior, scope, and what "success" looks like—that only surface after weeks of fieldwork. By then, sunk-cost bias makes them nearly impossible to revise.

Planned approach

An AI co-researcher adversarially reviews your plan: surfacing untested hypotheses, flagging leading questions, and proposing counter-scenarios you haven't considered. It won't write your plan—it will stress-test it.

View concept on GitHub
04

Insight Rigidity Checker

In development

Before sharing findings · prompt scaffold

The UX Problem

Research teams frequently "fall in love" with insights during synthesis—polishing narratives that feel compelling but aren't fully supported by the evidence. Stakeholders then act on fragile conclusions with high confidence.

Planned approach

For each insight you draft, the co-researcher will map it back to supporting quotes, flag overgeneralizations, assign a confidence score, and ask the question you might be avoiding: what would disprove this?

View concept on GitHub