Figure Eight combines human and machine intelligence to provide the highest quality training data in the industry, no matter what type of data you’re working with for your machine learning models.
Every Figure Eight job starts with data. Our platform leverages a pool of expert contributors and cutting-edge models to annotate your data for your specific needs. We accept .csv, .tsv, .xls, .xlsx and .ods formats and you can upload data either via our secure web portal or our RESTful API. We support text, image, URLs, videos, audio, and more. This is the data we’ll help you annotate throughout the rest of the process.
Figure Eight is a flexible platform that allows you to annotate your data your way. You can get started with templates for some of our most popular use cases like search relevance tuning or image annotation, then use our WYSIWYG editor to customize your Figure Eight workflow.
Because Figure Eight leverages human intelligence at scale, you’ll want to let our contributors know what you’re expecting. After all, every company has their own rules and needs for their data. In your instructions, we recommend spelling out step-by-step what you expect and giving concrete examples to achieve maximum quality.
Our patented test question feature is a key lever to how we get the highest quality training data for our customers. Essentially, test questions are a small subset of “gold” question-answer pairs that you answer and test contributors on their understanding on what precisely you need. Annotators need to reach an accuracy threshold you set to be allowed to annotate your data. And we weave in test questions as the job progresses so we can be sure quality remains high as work gets done.
You set the rules for your Figure Eight job. You can set your own accuracy thresholds, target,\ and segment for language or geography, customize workflows to send data from one job to others, optimize your settings for speed and cost, and more. Put simply: you create the rules that are right for your job. And we’re always here to help along the way if needed.
At this point, you’re ready to launch your job! Once you do, our global pool of contributors and tailored machine learning models will begin annotating your training data based on the instructions you provide to the quality specifications you want. You can monitor in real-time as your job progresses, allowing you to easily forecast when your job will finish, and even download results while the job is running.
The end result of this process is annotated data. Once your job has run, you’ll be able to download full and aggregated reports of every data row. You’ll see your original source data as you uploaded it, plus all the individual annotations and judgments our platforms provides. These reports are customizable after job completion and can be downloaded whenever you need them. Additionally, as jobs progress or finish on Figure Eight, you can send that annotated data to your own model via our customizable CSS.