Monitor your production AI model and data quality to improve real-world AI performance.
Production AI quality degrades over time due to drift. Radar monitors and sets configurable alerts on AI quality issues such as data drift, outlier data points, known errors, and more.
Traditional metrics such as accuracy can’t be measured in production without ground truth labels. Radar’s monitoring dashboard tracks model quality even without ground truth.
Radar automates operational monitoring and prediction logging of your AI systems to supports compliance against AI standards and regulations, such as the EU AI Act and ISO 42001.
Automatically monitor your production AI model and data quality to improve real-world AI performance.
Radar automatically captures your model’s prediction logs, allowing you to monitor individual predictions and aggregate trends such as data and model drift in real time.
Radar’s monitoring dashboard continuously detects and visualizes AI quality risks that cause real-world performance deterioration: data drift, prediction drift, data gaps, and more.
Radar’s built-in, customizable firewall blocks bad data points and bad model predictions. For example, enable a firewall filter to automatically flag outlier data points that are outside of the training data.
Radar automatically tracks the history of all model predictions in a searchable log – allowing you to easily audit predictions and meet compliance requirements.
Radar enables you to record, replay, and explain any model prediction on real customer data. Remotely generate prediction explanations for your model hosted on a remote server – no internal model access needed.
BSI is delighted to be partnering with Citadel AI to further enhance the reliability, safety and security of AI for the long-term benefit of society, organizations and individuals. Citadel AI have technology and skills that, combined with BSI’s technical knowledge and regulatory service expertise, can add real value in helping organizations to responsibly scale AI for the good of society.
Manuela Gazzard
Group Director, Regulatory Services at BSI
Let’s discuss how Citadel AI can improve your AI quality.
Track multiple models in production from a single dashboard.
Traditional metrics such as accuracy can’t be measured in production without ground truth labels.
Radar’s monitoring dashboard tracks model and data quality even without ground truth.
Radar automatically tracks the history of all model predictions in a searchable log.
Citadel Radar analyzes the input data and output predictions from your production AI server to automatically detect issues such as data drift, outlier data points, known errors, and more. Since Radar works without ground truth labels, it allows you to monitor your production AI quality in real-time.
When training an AI model, you usually use metrics such as accuracy to evaluate the model. However, these metrics can’t be used to evaluate production AI models since there’s no ground truth.
Instead, Radar monitors quality metrics such as:
These metrics enable you to track AI performance even without ground truth.
Of course, Radar can also retroactively track accuracy and other traditional metrics once ground truth labels become available.
To use Citadel Radar, you’ll need a model that you want to monitor. Radar can connect to your model in several ways: a real-time logging API, real-time firewall API, batch uploads, and more. During a trial, we can also provide a sample model and production data.
We’re happy to discuss the best setup for your application – contact us here anytime.
We take security seriously at Citadel AI, and we’re ISO 27001 certified and GDPR compliant. Your data is encrypted in transit and at rest, and your Radar server can be hosted in a cloud or on-prem environment of your choice. For more information, visit our Trust Report or contact security@citadel-ai.com.
We work with a wide range of users from startups to large enterprises, so our customers’ use cases and requirements differ significantly. Please contact us to determine the best pricing for your use case.
Interested in a product demo or discussing how Citadel AI can improve your AI quality? Please reach out to us here or by email.
