Citadel AI Inc., the leader in improving AI performance, has launched the beta version of Citadel Radar.

Citadel Radar provides automated monitoring for your data and model – improving real-world prediction quality and reducing maintenance time & cost. 

‍

‍

As society grows increasingly reliant on AI systems, concerns about AI accountability and bias are emerging in Europe and the United States, with the EU announcing proposed regulations on the use of artificial intelligence in April 2021. 

Without proper monitoring tools, AI systems become fragile black-boxes in your decision-making process, which are easy to break and difficult to fix. In order to build reliable AI systems, technology companies such as Google and Microsoft have published operational playbooks and ethical guidelines for AI.

After deployment, AI systems require constant attention to maintain real-world prediction quality while eliminating bias. However, for many companies, it's a significant challenge and cost for busy data scientists and software engineers to continuously maintain AI systems.

Citadel Radar automatically monitors enterprise AI systems, detecting, blocking, and visualizing a wide range of real-world risks (for example: data drift, invalid data, outliers), and can be applied to almost any ML model and application without requiring modifications.

Contact us on our website (https://www.citadel.co.jp) to see a demo of Citadel Radar.

‍

Common problems with real-world AI performance

• Performance gap between offline and real-world data
• Sudden or gradual performance drop due to data drift
• Bad predictions on low quality, outlier data points
• Hard to debug or understand black-box predictions
• Hard to A/B test new model versions
• and many more...

‍

Citadel Radar provides automated monitoring for your data and model

Monitoring

• Automated health alerts and analytics that detect performance degradation, data drift, and data gaps.

Firewall

• Improve AI performance by validating real-world data quality with hundreds of tests for invalid, outlier and adversarial data.

Explainability

• Explain any real-world prediction, of any model, at any time from local feature importance to counterfactuals.

‍

About Citadel AI Inc.

Citadel AI Inc. is a startup based in Tokyo. Our company recently received a seed grant from The University of Tokyo IPC, and is co-founded by the former product manager for ML infrastructure at Google Brain.

Citadel AI's vision is to provide you with AI that you can always trust. 

• Representative Director:    Hironori “Rick” Kobayashi
• Headquarters:                      Shibuya, Tokyo
• Establishment:                     December 10, 2020
• Company URL:                      https://www.citadel.co.jp