Good to see you here

Available for AI UX consulting

I'm Nelly, an AI UX consultant and product designer.

Most AI products don't fail on the algorithm.
They fail because nobody decided who's accountable when the machine is wrong. That's the part I design.

My numbers

2015 → now
0
years in product
2015 – now
0
products shipped
across 5 industries
0
book published
The Interface Is Not the System
NP
UX
ENG
DATA
PM
OPS

Teams I've partnered with across founders, product, engineering and data science; five companies, over a decade.

View experience

Latest work

2 of 6 shown

Currently exploring

Design radar
Authority

Graduated autonomy

I think every AI action needs a tier, not a toggle.

Trust

Confidence design

I believe sounding sure and being right are different problems.

Agents

Multi-agent conflict

I'm obsessed with what happens when two agents disagree.

Method

AI-native practice

I design with Claude, not just for it.

Governance

Risk as shared language

I think "high risk" is a negotiation, not a threshold.

The through line

"AI governance isn't a technical problem. It's a negotiation problem."

The system can't classify risk until the organization agrees on what it values. That agreement is the work.

My Work

Six products across enterprise AI, fintech, civic tech and document automation.
Each shipped with a real constraint: a regulator, a deadline, or a team that didn't agree yet.

Case studies

07 projects

My Process

I don't believe in one-size-fits-all.
Every project gets the approach it needs. Sometimes deep research, sometimes shipping code in a day.

What kind of engagement?

About Me

The short version, the working version, and what I do when I've been staring at systems too long.

AI UX consultant Nelly Pourmehr
Hi, I'm Nelly
A short letter about me

Ten years in product design, most of it on AI and SaaS products where the hard part was never the interface. I'm an AI UX consultant now, but the question hasn't changed: what happens when the system is wrong, and who answers for it?

I work with product, design, and engineering teams to build AI systems that can be trusted, questioned, and governed. That means asking the uncomfortable question early, before it becomes a postmortem.

I wrote a book about it, The Interface Is Not the System, and I run those same frameworks on real client work. Based in Germany, working with teams anywhere.

Thanks for being here

inConnect on LinkedIn
Working principles

What I believe

What it's like

Working with me

Outside the work

When I'm not designing

Paintings, places, and things I cook. Hover to pause · tap any tile to open.

Experience

Ten years, mostly where AI meets accountability. Open a role to see the detail.

Roles

Reverse chronological

Beyond the projects

Reach
Mentoring

Helping UX designers adopt AI into their own practice, from prompting and evaluation to where human judgment still has to lead.

AI consulting

Runs Groundwork, a six-check AI trust diagnostic: from use case readiness to launch risk to live agent grounding. Evidence labeled, sourced, no black box.

Writing

Author of The Interface Is Not the System. Ongoing writing on AI UX at uxandai.com.

Download resume

Writing

One book and an ongoing argument about AI, design and who is accountable.
Pick a shelf.

PublishedAI UXFramework

The Interface Is Not the System

Why most AI products fail at the seam between the model and the person using it, and the framework I use to fix it.

The premise

Fixing the screen rarely fixes the problem.

Teams pour effort into the chat window, the loading state, the empty screen. Meanwhile the actual failure is upstream: nobody decided what the AI is allowed to do, how sure it needs to be, or who is answerable when it's wrong. The interface only makes those decisions visible. It doesn't make them.

Part 1

Thinking Before Designing AI

Framing the problem correctly before any interface gets drawn.

Part 2

Core Design Patterns for AI Systems

The recurring patterns that make AI behavior legible instead of opaque.

Part 3

Designing for Uncertainty, Time, and Change

Confidence, drift, and what happens after ship day.

Part 4

Research, Bias, and Ethics in AI Product Design

Where responsibility actually sits, and how to research for it.

The six failure patterns

Part one
Estimated prevalence of each anti-pattern across the products I reviewed
Blame deflection90% – 95%

Legal disclaimers ("AI may make mistakes") combined with designs that encourage high reliance.

False confidence75% – 85%

Fluent, authoritative text generation without visible calibration or uncertainty indicators.

Invisible scope70% – 80%

Unclear context windows, hidden system prompts, and ambiguous capabilities.

Unowned decisions60% – 75%

Agentic tools taking implicit actions without clear attribution or explicit confirmation.

Rubber-stamp review55% – 70%

Oversight screens designed to cause fatigue, leading users to auto-approve AI output.

No undo path50% – 65%

Destructive or overwriting actions (code refactoring, content replacement) lacking version history.

Inside the book

Spreads
Who it's for

Anyone shipping AI into a place where being wrong is expensive.

Product designers, PMs, and the engineering leads who end up owning the governance conversation by default. Not a technical book: no model architecture, no code. It's about the decisions around the model.

Need a workshop for your team?

My AI Design Tools

Tools, skills and maps I build for my own practice.
Not client work. This is where I test ideas before they earn a case study.

Live
0

Built and usable right now

Coming soon
0

In build, shipping soon

In planning
0

Scoped, not started

Available for consulting

Let's talk about what you're building

Open to senior product roles, consulting engagements and AI product audits. If the design isn't decided yet, that's the best time to talk.

Send a message
NPWhen I'm not designing