A code-free workspace for knowledge graphs

Client
Stardog
Timeline
2019 to 2020 · 9 months
Role
Senior UX Designer · Team Lead
Team
Modassir Iqbal (co-lead), Natalie Yeh, Yirang Choe, Monikka Ravichandran, Aravind Jembu Rajkumar

Impact

Faster time-to-graph on the auto-mapping flow
25%
New users in the months after shipping
1,000+
Audiences served by one workspace
3 → 1
Stardog Studio knowledge graph workspace mockup

02Background

A platform powering NASA and Fortune 500s, gated behind code.

Stardog's knowledge graph platform sits underneath NASA, Bank of America, CapitalOne and Dow Jones. Defining a graph, and mapping source data into it, meant writing SPARQL by hand.

03Problem

The people who own the data's meaning couldn't shape the model.

The brief: let analysts and SMEs build graphs without code, and keep SPARQL intact for the engineers who relied on it.

Lead ontologist

High-tech

Five-plus years in knowledge graphs. Lives in SPARQL; the code stays the source of truth.

Junior data architect

Low-tech

Comfortable in code but slow at it for graphs. Wants the canvas to absorb the boilerplate.

SME / business analyst

Non-tech

Holds the business meaning of the data. Has never written SPARQL and shouldn't have to.

04Research

11 interviews showed modeling and mapping happen in one loop.

Semi-structured interviews
11
Contextual inquiries on real datasets
5
Observations on the affinity wall
200+
  1. Modeling and mapping are one loop

    Users added a class, found the data needed a property, then went back to the model.

  2. High-techs type faster than they click

    Code mode had to stay first-class, and in sync with the visual canvas.

  3. Auto-mapping is the real starting point

    Most active users began new projects from auto-mapping rather than a blank canvas.

  4. Properties come before classes

    Users defined a pool of properties first, then assigned them to classes.

Affinity wall

05System

Auto-mapping first, one synced canvas, and a Blueprint.js component set.

Each move went through lo-fi tests with 5 participants, then 2 A/B rounds with 8 more.

  1. Auto-mapping first

    Project creation defaults to auto-mapping. Naming the project comes after the work.

  2. One canvas for both jobs

    Modeling and mapping share a canvas, with a synced code and visual switcher for high-techs.

  3. Properties on their own

    A property drawer lets users define properties first, then assign them to any class.

Auto-mapping first

1 of 4

Unified canvas

2 of 4

Flexible properties

3 of 4

Design system on Blueprint.js

4 of 4

06Results

Stardog shipped the workflows and components into Stardog Studio.

Faster time-to-graph on the auto-mapping flow
25%
New users in the months after shipping
1,000+
Audiences served by one workspace
3 → 1

07Reflection

Leading the team taught me as much as the design did.

  1. Learning the domain paid off most

    Weeks of pair-programming with Stardog's engineers, before we drew anything, shaped every later visual decision.

  2. I'd test lo-fi sooner

    Findings were stable by week six. Prototypes in users' hands earlier would have shortened the next two months.

  3. Leading in a new domain

    Coordinating six designers, two researchers and an SME on unfamiliar ground pushed the project as hard as the design.

Imi

Building Imi, an AI nutrition coach, from brand to React Native front end

2026, 10 min read

IntuitLocked

AI Chart of Accounts cleanup that accountants now finish, from ~3% to ~9%

2025, 7 min read

IntuitLocked

An AI reporting layer for mismatched charts of accounts, books untouched

2026, 6 min read

Rocket Lawyer

Rocket Lawyer's first design system, shipping marketing pages 14× faster

2022, 7 min read

DocOn Technologies

Live wait times for patients in 700+ hospitals, on a TV app and a doctor dashboard

2017, 7 min read

University of Maryland

A new brand and site for Lift Academy, with enrollment up 50% month over month

2018, 6 min read

Home