Clients can interact with Uber Data Labeling services at any stage of labeling, from request submission to download results. Below are a few sample interactions that clients can have.

• To provide data labeling artifacts(datasets) for labeling.
• To download annotated labels after labeling work completion.
• To track the labeling status at the request level, ingestion level, etc.,
• Create new workspaces/projects and add the labeling configurations, cloud integration details, etc.,

Uber’s Data Labeling services offer multiple integration modes for clients to support the above interactions as per their varied use cases. Below are the primary integration mechanisms.

1. [Signed URL] Clients can provide access to their raw data through signed URLs. Export of labeled data (output) can also be supported through signed URLs.
2. [Cloud-based] Automated ingestion of data to be labeled and Retrieval/Export of labeled data via Cloud storage solutions.
3. [API-based] Rich API-based programmatic integration.
4. [Hybrid] Flexible hybrid integrations that combine #1 and #2. For example,
   • Clients can host input artifacts in S3, submit API requests for ingesting stored artifacts(input location), poll API for status checks, and configure cloud-based results upload to S3.
   • Clients can host input artifacts in S3, configure cloud-based scheduled ingestion from S3, poll API for status checks, and use API-based on-demand downloads.
5. [Customized] Uber will also be able to support a tailor-made integration mechanism based on client requirements.

¶ Automated Data Ingestion & Delivery

Uber supports automated data ingestion from cloud storage platforms like GCP GCS, AWS S3, and Azure Blob Storage. We follow industry best practices to ensure secure data interchange between Uber and client’s infrastructures.

An integration between Uber and your cloud infrastructure is necessary to enable automated data ingestion and delivery. For detailed step-by-step instructions on setting up cloud integrations, please refer to the Data Hosting page.