Why Giskard?

Giskard is a holistic Testing platform for AI models to control all 3 types of AI risks: Quality, Security & Compliance.

It addresses the following challenges in AI testing:

  • Edge cases in AI are domain-specific and often seemingly infinite.

  • The AI development process is an experimental, trial-and-error process where quality KPIs are multi-dimensional.

  • Generative AI introduces new security vulnerabilities which requires constant vigilance and adversarial red-teaming.

  • AI compliance with new regulations necessitate that data scientists write extensive documentation.

Giskard provides a platform for testing all AI models, from tabular ML to LLMs. This enables AI teams to:

  1. Reduce AI risks by enhancing the test coverage on quality & security dimensions.

  2. Save time by automating testing, evaluation and debugging processes.

  3. Automate compliance with the EU AI Act and upcoming AI regulations & standards.

Giskard Library (open-source)

An open-source library to scan your AI models for vulnerabilities and generate test suites automatically to aid in the Quality & Security evaluation process of ML models and LLMs.

Testing Machine Learning applications can be tedious. Since AI models depend on data, quality testing scenarios depend on domain specificities and are often infinite. Besides, detecting security vulnerabilities on LLM applications requires specialized knowledge that most AI teams don’t possess.

To help you solve these challenges, Giskard library helps to:

  • Scan your model to find hidden vulnerabilities automatically: The giskard scan automatically detects vulnerabilities such as performance bias, hallucination, prompt injection, data leakage, spurious correlation, overconfidence, etc.

  • Instantaneously generate domain-specific tests: giskard automatically generates relevant, customizable tests based on the vulnerabilities detected in the scan.

    ../_images/test_suite_scan_llm.png

  • Integrate and automate testing of AI models in CI/CD pipelines by leveraging native giskard integrations.

    ../_images/gh_discussion.png

Get started now with our quickstart notebooks! ⚡️

Giskard Hub

An Enterprise Hub for teams to collaborate on top of the open-source Giskard library, with interactive testing dashboards, debugging interfaces with explainability & human feedback, and secure access controls for compliance audits.

  • 🔍 Debug your issues by inspecting the failing examples of your tests (⬇️ see below the DEBUG button)

  • 📖 Leverage the Quality Assurance best practices of the most advanced ML teams with a centralized catalog of tests

  • 💡 Create hundreds of domain-specific tests thanks to automated model insights (⬇️ see below the bulbs 💡).

  • 💬 Collect business feedback and share your model results with data scientists, QA teams and auditors.

Get started now with our demo HuggingFace Space or by following installation instructions! 🐢