Release Highlights

General

The following updates are included in AI Developer Edition 1.2.0.

• Added badges to the README for improved visibility and quick access to key resources.
• Restructured directory for better organization of samples and source code. The files for each feature are now available in the feature directory.

Data Discovery

Data Discovery provides the functionality to explore and analyze datasets for sensitive information. For more information about Data Discovery, refer to the feature section.

• Upgraded to Data Discovery version 2.0.
• Added direct-API example scripts for text classification, tabular (CSV) classification, and redaction, both Python and bash variants.
• Introduced an isolated data-discovery/docker-compose.yml for starting only the Data Discovery service.
• Updated API endpoints to the v2 paths: v2/classify/text, v2/classify/csv, and v2/transform/label.

Semantic Guardrails

Semantic Guardrails provides functionality to enforce policies and guidelines within AI interactions. For more information about Semantic Guardrails, refer to the feature section.

• Expanded domain model coverage to three verticals: customer service, financial, and healthcare.
• Extended the pii processor to user messages, in addition to AI messages, enabling PII detection on both sides of a conversation.
• Improved privacy in API responses: PII explanation output now returns character spans instead of the actual detected PII values.
• Changed the OpenAPI documentation endpoint from /swagger to /doc.
• Added a Jupyter notebook sample for seamless evaluation and execution.
• Included richer examples in the sample files for easier understanding.

Synthetic Data

Synthetic Data provides functionality to generate artificial datasets that mimic real data while preserving privacy. For more information about Synthetic Data, refer to the feature section.

• Expanded generative model support. In addition to GANs, Tabular Variational Autoencoders (TVAE) and diffusion-based models are now explicitly supported.
• Added the typeHint parameter to the generate request payload, allowing explicit selection of the model type. For example, "model_type": "tabdiff" for diffusion-based generation.
• When typeHint is not specified, the system automatically determines the most appropriate model during training.
• Refactored the synthetic data generation code for improved performance and maintainability.
• Updated the Jupyter notebook samples for quick evaluation and execution.

Anonymization

Anonymization provides functionality to remove or obscure personally identifiable information (PII) from datasets. For more information about Anonymization, refer to the feature section.

• Added new feature for the AI Developer Edition: Anonymization.
• Included sample code and documentation for using the Anonymization feature effectively.

Data Protection API Wrappers

Data Protection API Wrappers provide functionality to interact with the Data Protection APIs for tokenization, masking, and other privacy-preserving operations. For more information about Data Protection API Wrappers, refer to the feature section.

• Protegrity Data Protection Jupyter notebook to quickly test tokenization.
• Provided support and sample implementations for both Python and Java.
• Ensured Java samples are fully compatible across Linux, macOS, and Windows.
• Delivered Java source code for customization and compilation flexibility.