Lena Lehuby

Product owner · UX and applied AI · Paris

From user research to working software.

I defined the scope, chose the approach, set the delivery order and agreed with the team how to validate the result, starting from user research and prototypes. My experience covers B2B and B2C products.

Background in product design, and an MSc in Applied Data Science and AI.

Line drawing of Lena Lehuby

Selected work

Four projects, two or three decisions each

For each one: the problem, what I contributed, the decisions that mattered, and where the work stands. Screens come from my own prototypes. Where a client's material is confidential, the case stays at the level of the approach.

01 Yexit · AI feature · 2025–now Capturing a person for Portrait AI, in their own words AI product owner and product designer. A personality model built on McAdams and the IPIP, Profile insights computed without AI, and Portrait AI planned for after the answers are released, following the workflow of Park et al., with no fine-tuning. Status Spec, wireframes and user tests. Portrait AI is not built yet. 02 Stellantis · R&D internship · 2026 Traffic that moves naturally in VR, with the physics baked in R&D intern, development and project management. Live physics was too heavy for VR, and cars pushed along a path glided. I ran the physics once, baked it into the scene as USD animation and replayed it in VR. A scenario became a route, a speed and where to stop. Status Tools completed and validated with the team 03 Yexit · Design system · 2026 An AI-ready design system for Yexit AI product owner and product designer. One reference for people and AI tools: three token layers, a usage note on every colour, rules written as constraints, components in code. Status Default system for AI-generated Yexit screens 04 Own project · Planning tool · 2026 Sprint tool Own project, built with AI. One page that estimates what an AI feature costs to run, the hours it takes to keep alive, and what that leaves for delivery. No login, nothing leaves the browser. Status First release, live on this site
05

AI workflow · Managing AI features

An AI feature is not finished at launch

How I manage AI features: a delivery loop and a reliability loop planned from the start, seven decisions for every feature, eight conditions before launch, and the same rules applied to my own work.

Read the AI workflow →

Method

How I work

R1

Documents go out before the meeting. The meeting is for deciding.

Yexit kickoff brief
R2

Every open decision leaves the room with an owner and a date. The value can come later.

Yexit decision log
R3

Failure states are in scope from the start: offline, interrupted, declined, empty, denied.

Yexit prototypes
R4

In a small team, the user test runs every week and the backlog comes out of it.

Mindii
R5

Documentation describes what the code does. If the help text and the code disagree, the text says so.

Stellantis user guides
R6

Trade-offs are made explicit as quality, time, cost and confidentiality, and decided in that light.

Yexit
How I manage AI features →

Working with me

Three things to count on, with the evidence

Range

Audiovisual production, then UX and product design in B2B SaaS, then product ownership, then an MSc in applied AI. At Yexit the same person wrote the user flows, the acceptance criteria and the editorial half of the model prompt. I can read a spec, a Figma file and a Python module, and write all three.

Work

The two Stellantis extensions were built alone in six months, alongside the MSc thesis, and each has its own user guide. At Yexit, the specification and UX guide gave the developers the flows and acceptance criteria to build from.

Patience

The Yexit family tree went through ten rounds of screenshots and ended as written rules: each connector has one meaning, and a fixed order settles conflicts. At Mindii, a user test ran every week for months, and its findings set what the team fixed next.

Contact

Product owner and product manager roles in B2B SaaS and AI.

Email hello@lenalehuby.com
Based in France
Lena Lehuby · 2026 Screens from my own prototypes · sample data throughout