Mondrian AI · Yennefer

From an IDE proof of concept to a connected enterprise AI product

I led product design for Yennefer from its first research to launch, expanding it from one IDE into three connected applications for AI researchers and administrators.

Role

Founding product designer

Timeline

12 months, concept to launch

Team

Founders, 4 engineers, no PM

Scope

Enterprise AI and MLOps

app.mondrian.ai / studio

The shipped Yennefer Studio home, organized around projects

ShippedStudio home. Projects and their status come first.

Problem Direction Data Catalog Project home Impact

Problem, role, research

AI developers were working across fragmented tools

I joined when Yennefer was an IDE proof of concept. Engineering had validated the infrastructure, but not who the product was for or where it should start and end.

With no PM or researcher on the team, I led the research: 10+ user interviews, 10+ domain reports and 20+ competing products.

Most of the work happened before the IDE opened. One AI engineer spent about two weeks collecting usable data from colleagues, shared drives and downloads before an experiment could start.

InheritedThe IDE proof of concept. It opened on server state, CPU, memory and disk.

Discover

Outside

Evaluate

Outside

Prepare

Outside

Build

In the IDE

Run

In the IDE

Monitor

In the IDE

Outside

The research lifecycle mapped from interviews. Four of seven steps happened in shared drives, chat and email.

Product direction

The product expanded beyond the IDE

I recommended moving the product boundary upstream to data discovery and adding a governance surface for administrators. One IDE became three applications on a shared model of projects, datasets, permissions and compute.

This added scope and engineering effort. It also meant solving the full workflow instead of only the middle of it.

Before · IDE

Build→Run→Monitor

After · Platform

Discover→Build→Run→Monitor→Govern

Data CatalogDiscover

Data Catalog dataset page with source, publisher, update date and tags

StudioBuild

AdminGovern

Admin dashboard with resource gauges, usage trend, GPU table and project table

Design decision 01 · Data Catalog

Bring data discovery into the product

Problem

Researchers found, checked and downloaded data outside the platform, then rebuilt that context inside the IDE. A standalone repository would have kept that gap.

Decision

I designed the Data Catalog as the first step of the workflow: find a dataset, check its source, license and freshness, and import it directly into a Studio project.

Why

Licensing and data quality decided whether a dataset was usable, so they had to be visible before a researcher committed time. Direct import removed the download and re-upload step between discovery and experimentation.

Data Catalog search across indexed sources, filtered by organization and topic

FindSearch across indexed public and organization sources, seeded so the catalog was useful on day one.

Import a dataset into an existing Studio project with a project selector

ImportDiscovery ends inside a Studio project, not in a download folder.

Design decision 02 · Project-first home

Make the project the home screen

Problem

The IDE home was organized around servers, CPU, memory and disk. To resume work, researchers had to remember which machine their project lived on.

Decision

I made the project the primary object. The home shows project status, owner, collaborators and last activity, with infrastructure moved into supporting context.

Why

In usability testing, researchers asked to see project status first. They returned to continue a project, not to operate a server.

app.mondrian.ai / studio

ShippedProject status, ownership and collaborators before anything about infrastructure.

“I want to see overall projects’ status and latest projects, rather than data and AI model resources, on the main page.”
AI researcher · usability testing

Connected system and impact

One product model across three applications

A dataset imported from the Catalog appears in a Studio project, and that project’s GPU usage appears in Admin. Administrators allocate capacity by project instead of by person.

The product launched in 12 months and was adopted by enterprise and education organizations.

Dataset→Studio project→Compute usage→Admin allocation