Scale Data Engine
Product · API / developer platform by Scale AI
Collect, curate, and annotate data. Train models and evaluate. Repeat.
Scale Data Engine is a data labeling and curation platform for businesses and governments to build high-quality training datasets. It combines ML pre-labeling with human expert review to support text, image, video, and 3D sensor fusion data at production scale.
- Languages
- English, Arabic
Category: Data Labeling
- APIs and SDKs
- REST API and endpoint reference documentation
- Supported Sensors
- LiDAR, RADAR, Electro-Optical, Infrared
- Data Formats
- Text, Image, Video, 3D Point Cloud
- Cloud Integrations
- AWS S3, Google Cloud Storage, Azure Blob Storage
- Team Roles
- Admin, Manager, Member, Labeler
- Human Decisions
- 15B+ for model training
- 15B+ human decisions to train AI models
- Supports LiDAR, RADAR, and 3D Sensor Fusion
- Integrated with AWS S3, Google Cloud, and Azure
- Expert-led RLHF and model evaluation
Overview
Scale Data Engine is a comprehensive platform for collecting, curating, and annotating data to train and evaluate machine learning models. It serves businesses and government agencies by turning raw data into high-quality training sets through a combination of machine learning pre-labeling and human review from a global network of experts. The platform supports diverse data types including text, image, video, and 3D sensor fusion like LiDAR and RADAR for autonomous systems. By integrating domain experts and automated quality control, it helps teams find and fix model failures while optimizing labeling costs. Scale Data Engine is designed to scale from small experiments to massive production-level AI programs, with over 15 billion human decisions already processed.
Supported Data Types
- Scale Text: The platform supports document processing, natural language processing, transcription, and content analysis for text-based datasets.
- Scale Image: Scale handles electro-optical and infrared image data, including transcription for static visual assets.
- Scale Video: The engine processes full-motion video and provides natural language processing for temporal data.
- Scale 3D Sensor Fusion: Specialized annotation for LiDAR and RADAR data supports the development of autonomous systems and robotics.
Data Curation & Annotation
- Expert Labeling: Scale provides high-quality labels from a global network of hand-picked subject matter experts across diverse fields.
- ML-Powered Pre-labeling: The platform combines machine learning powered pre-labeling and active tooling with varying levels of human review.
- Failure Analysis: Users can easily find, categorize, and fix model failures to optimize labeling spend on high-value curated data.
- Elastic Scalability: The engine supports any ML project from lower-volume experiments to high-volume production workloads.
Developer & Team Tools
- API Access: Scale provides a dedicated API for programmatic access to concepts and endpoint reference documentation.
- Cloud Data Hosting: The platform integrates with AWS S3, Google Cloud Storage, and Azure Blob Storage for data hosting.
- Management Dashboard: The Customer Dashboard allows teams to manage accounts, tasks, projects, and quality metrics in one place.
- Role-Based Access Control: Admins can invite others to join teams as Managers, Members, or Labelers with specific action permissions.
Use cases
- Autonomous vehicle teams use 3D sensor fusion to annotate LiDAR and RADAR data for navigation models.
- Generative AI developers use RLHF and human feedback to align model outputs with human preferences.
- Healthcare organizations process medical records and clinical history to train diagnostic and reasoning agents.
- Public sector agencies use the platform to fine-tune computer vision algorithms for perimeter security and mission planning.
Specifications
- APIs and SDKs
- REST API and endpoint reference documentation
- Supported Sensors
- LiDAR, RADAR, Electro-Optical, Infrared
- Data Formats
- Text, Image, Video, 3D Point Cloud
- Cloud Integrations
- AWS S3, Google Cloud Storage, Azure Blob Storage
- Team Roles
- Admin, Manager, Member, Labeler
- Human Decisions
- 15B+ for model training
- Access Method
- REST API and Web Dashboard
How it works
- Connect and Upload Data: Connect to your data sources in AWS S3, Google Cloud Storage, or Azure Blob Storage and upload your raw data artifacts.
- Configure Projects and Tasks: Use the Scale API or dashboard to set up projects, define taxonomies, and select the type of human or ML review needed.
- Annotate and Curate: Scale's engine applies ML pre-labeling and human expert review to annotate and curate the datasets.
- Evaluate and Export: Monitor project metrics and quality tasks through the dashboard, then export the high-quality labeled tasks for model training.
FAQ
What types of data can Scale Data Engine process?
Scale supports a wide range of data types including text (NLP, transcription), image (electro-optical, infrared), video (full-motion), and 3D sensor fusion (LiDAR, RADAR).
How does Scale ensure data quality?
Scale combines machine learning powered pre-labeling with varying levels and types of human review from a global network of vetted subject matter experts.
Which cloud storage platforms are supported?
Scale supports data hosting and integration with major cloud providers including AWS S3, Google Cloud Storage, and Azure Blob Storage.
What team roles are available in the platform?
Scale accounts support multiple roles including Admin, Manager, Member, and Labeler, each with specific permissions for managing projects and data.
Added by AntSpace from public sources (scale.com).