Quality assurance · Production

Labelbox accelerated the launch of text to image models for a frontier AI lab

The problem

An AGI research company preparing to launch an AI product needed rapid content moderation at scale while avoiding lock-in to a single labeling vendor, which had historically caused cost overruns, delays, and inability to flexibly scale. Their existing internal labeling tool was slow and resource-intensive to develop and struggled to keep up with expanding use cases.

First attempt

The company's previous internal labeling tool was slow and engineering-intensive to build and maintain, and prior experiences with external vendors involved significant delays when starting labeling projects.

Workflow diagram · grounded in source
1
Product launch triggers labeling need
trigger
“The company was planning on launching a highly anticipated AI product and needed to deliver a solution for rapid content moderation”
2
Multi-vendor labeling platform setup
integration
“the company leveraged Labelbox's platform which provided the ability to test and contract multiple labeling vendors simultaneously. This allowed them to rapidly manage over 10 labeling vendors - from initial introduction to project kick off”
3
Project configuration and data import
integration
“the company was able to configure projects in a matter of minutes for their primary data types which included both images and text. In addition, the company utilized Labelbox's webhooks and attachments to make the process of data import …”
4
Human safety content annotation
human_review
“review and generate hundreds of thousands of annotations in the span of just three months for their content moderation use case (which focused on tagging safe vs. unsafe content)”
5
Real-time flagging and model refresh
feedback_loop
“the company integrated a real-time pipeline for flagging and reporting images with the ability to refresh their existing models. This accelerated the speed of removing unsafe content in the form of images that were violent, toxic, lewd, …”
Reported outcome

Using Labelbox, the company managed over 10 labeling vendors simultaneously and generated hundreds of thousands of annotations in three months, enabling a hugely successful AI product launch with widespread industry excitement and user adoption.

Reported metrics
Labeling vendors managedover 10
Annotations generated in three monthshundreds of thousands of annotated assets
Project configuration timematter of minutes
Product launch outcomehugely successful product launch
Reported stack
Labelboxwebhooks
Source
https://labelbox.com/customers/AGI-case-study/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Using Labelbox, the company managed over 10 labeling vendors simultaneously and generated hundreds of thousands of annotations in three months, enabling a hugely successful AI product launch with widespread industry e…

What tools did this team use?

Labelbox, webhooks.

What results were reported?

Labeling vendors managed: over 10; Annotations generated in three months: hundreds of thousands of annotated assets; Project configuration time: matter of minutes; Product launch outcome: hugely successful product launch (source-reported, not independently verified).

What failed first in this deployment?

The company's previous internal labeling tool was slow and engineering-intensive to build and maintain, and prior experiences with external vendors involved significant delays when starting labeling projects.

How is this quality assurance AI workflow structured?

Product launch triggers labeling need → Multi-vendor labeling platform setup → Project configuration and data import → Human safety content annotation → Real-time flagging and model refresh.