Improving project setup for Mighty Ai

I led UX efforts to improve project creation and throughput at Mighty Ai, where we provide training data for autonomous vehicles and other AI missions

The challenge

Mighty Ai’s project setup involves configuring objects and data in JSON format for every task. Setting up this file is a time-consuming process that requires a good bit of technical knowledge, which creates a roadblock for new team members, and also wreaks havoc on those of us already familiar with the system (myself included) — we tend to make mistakes because it’s so easy to miss them in a code editor.

I set out to research and prototype a new idea with a goal of reducing errors and speeding up our project kickoffs.

Job JSON Editor wireframe

Project goals

  1. Identify the minimum requirements for setting up a task
  2. Create a wizard workflow to guide us through project creation
  3. Significantly reduce errors and kick off projects more quickly
  4. Build a JSON preview and provide advanced options for power users

Conducting research

I first catalogued all of the available JSON configuration options. Many were stale; some were used in every project. Research conducted with the customer success team revealed the settings they used most often, and with those findings in hand, I had an idea of what decisions the system could make on its own. I set off to map out some user flows.

Simplifying the setup

We needed a logical way to walk people through the setup, which required rethinking our process since we were used to doing it all at once. It soon became obvious that we really had to create a simple way to associate classes with their annotation type(s) and potential metadata. A tabular layout with tooltips worked best to expose the options without overwhelming people.

Early sketches of the JSON editor

Viewing the output

After inputting annotation types, classes, and metadata, the last step of the wizard showed a preview of the JSON file built according to your choices. If you needed advanced functionality, you could make adjustments directly in the file. Changes would be reflected in each step of the wizard if you needed to go back and edit an earlier panel.

JSON editor mockups

Testing with users

After creating user journeys, wireframes, and mockups, I built a prototype in HTML, CSS, and JavaScript to test with the customer success team and product management.

This feature has garnered praise for its impact across onboarding, revenue, and engineering, and is currently in queue for development.

JSON editor class creator
EvoLux