© Copyright 2026 | by david nguyen
Experiments in AI for design and research

The Lab is where AI experiments meet real people.

I design and build small experiments that ask one question: what can AI actually do for design and research? Each one gets tested with real people before I publish anything. Most stay experiments. One became User Sage, the user research and usability testing platform I run today.

Questions? Ask the virtual one. The real one is tinkering in the Lab.

See the builds
Line drawing of David Nguyen
Ask David (the Virtual one)
Answers about the Lab & the work
Not alive
Hi. I am the AI version of David, and this is his Lab. Everything below is one of his experiments. Ask me what this place is, what ProductRubric does, or which one you should try first.

I'm the fast David, not the careful one. Double-check anything important with the real one.

5 experiments · User Sage left the bench AI product design · Evaluation · Research tooling · Prototyping
EXP.1 EST. 2026Shipped · Oct 2026

ProductRubric

Evaluation · Open source · MCP

Every AI critique I saw was improvised on the spot. Ask the same thing twice and you got two different standards. I wanted the standard written down, once, where any AI tool could pick it up and run it the same way every time.

Homepage review, recorded
Design Principles Score · 72 / 100

This is a recording of an actual chat. The ask was one plain sentence, and the assistant picked the rubric and scored the page by itself. The full recordings live on productrubric.com.

Views · Swipe or use the controls

View 1 · The library

20 rubrics, 4 fields

The library has seven for product design, four for product research, five for product management, and four for product marketing. Every one uses the same format, so any AI assistant can run any of them.

View 2 · A rubric, up close

Each rubric is one file: the criteria, the scoring scale, the exact shape of the report, and honest notes on what the rubric cannot judge. That file is the whole product.

View 3 · Proof it runs

The Survey Review worked example in the repository was produced by the project's own MCP server, and it matched the answer key exactly.

My design review tool, Design Sage, had its evaluation criteria locked inside one product. Teams could not bring their own standards, and nothing carried from one AI tool to another. ProductRubric takes the criteria out of the tool and publishes them as files. Anyone can use them, edit them, or write their own.

What it does +
ProductRubric is an open library of evaluation rubrics that an AI assistant can run against product work. Each rubric writes down the criteria, the scoring, and the shape of the report in advance, so a design, a survey, or a set of findings gets judged against the same standard every time. The library launched in October 2026 with ten rubrics across product design and product research, and grew within the week to twenty rubrics across four fields. You can take them as plain files, install them as Skills, or call them over an MCP server. It is free and open source.
Visit productrubric.com →
EXP.2 EST. 2025Live product

Design Sage

AI design review · Figma

A designer can wait a week for a crit and still get feedback that depends on who happened to be in the room. I wanted every designer to be able to get a structured first review in minutes, before the work ever reaches the team.

The plugin at work

This is the plugin working on a dashboard. On its own it cycles from the color palette to the layout variations to the attention heatmap, and it explains why users look where they look.

I have run design reviews for years, and the same principles decide most of them. Writing those principles down carefully enough for an AI to apply them turned out to be the hard part, and it is the part I am most proud of here. The principles started as my own review practice and became the product.

What it does +
Design Sage is an AI design review tool that lives inside Figma. It reviews a design against a set of design principles and returns scored feedback, written critique, and attention heatmaps. Some of its principles now also live in the ProductRubric library, where any AI assistant can run them.
Visit designsage.app →
EXP.3 EST. 2026Live · In Grok

User Sage Grok Bot

Research tooling · Grok

Researchers and designers already spend their day in chat. I wanted them to be able to plan and launch a real usability study from the conversation they were already having, instead of moving to another tool and starting over.

Designer and researcher, in chat

Plan a first-click test for our new pricing page.

Drafting study · 5 tasks · success criteria set

Study ready. Run an AI panel of 12 for a first read, or recruit real participants from Your Panel.

Panel result: 9 of 12 found Pricing on the first click. Hypothesis logged, ready to confirm with real people.

The characters come from the bot's launch post. The bot does two kinds of work in chat. It critiques designs and scores them, and it can build a study, run it, recruit participants, and analyze the findings.

User Sage is the main product at Chaotic Domain, my company, and the bot is its front door inside Grok. It also keeps me honest about my own rule for AI in research. An AI panel gives you a hypothesis and tells you where to look next. Findings take real people. The bot offers the simulated read and the path to real participants in the same place.

What it does +
The User Sage Grok Bot brings User Sage into Grok. From a conversation, a researcher or designer can plan a usability study, build it, and run it, including running an AI panel for a first read before testing with real people. It can answer study design questions on its own. When someone wants to run a real study from the chat, it connects to their User Sage workspace.
Visit usersage.com →
EXP.4 EST. 2025Live product

Sleuth AI

AI assistant · Slack

Decisions happen in Slack threads, and then the thread scrolls away and the follow-up never happens. I kept watching good calls get buried, so I helped build the thing that watches the thread for the team.

Asked in Slack, answered in thread

This happened in #team-ux. The question was typed in the channel like any other message, and the answer came back in the thread: #CDCDCD on white is 1.07 to 1, which does not meet the minimum contrast for accessibility.

Every team I have worked with loses the same things in Slack: the decision, the deadline someone mentioned in passing, and the summary nobody had time to write. Sleuth AI lives where the conversation already is and does that follow-through work, so nobody has to become the team's note taker.

What it does +
Sleuth AI is an AI assistant for Slack. It listens in your channels, answers questions in place, creates automatic reminders from what people say, and produces instant summaries of project channels. It can also turn messages into tasks and assign them, and flag project risks it spots in team discussions. Pricing is per organization rather than per seat.
Visit sleuth-ai.app →

The one that left the bench

Started on this bench · Now a full product
USERSAGE.COM

User Sage

User Sage is a user research and usability testing platform. It started as an experiment on this bench. Now it runs the whole job end to end: a team builds the study, recruits participants or runs an AI panel for a first read, schedules and holds live moderated sessions while stakeholders watch from the observer lounge, and pays participants through the built-in Tremendous integration, with every finding kept in one project.

AI gives you the hypothesis. Findings take real people.

Study builderAI PanelLive moderated sessionsObserver LoungeTremendous incentivesYour PanelMCP connector
Visit usersage.com
usersage.com
Live · usersage.com/lab, embedded hereThe User Sage booking page for a live session, showing the calendar and time slots a participant picks fromA prototype test result in User Sage: a third of the panel reached the target screen, and the verdict is Needs workA preference test result in User Sage: Option B leading with 11 of 12 votes, the margin, and the confidence noteBooking page · Live sessionsPrototype test · ResultsPreference test · Results
New builds are added as they shipTested with real people, like everything here