We’re excited to announce Algorithmia Insights, a flexible integration solution for ML model performance monitoring. Algorithmia Insights provides a pipeline of algorithm inference and operational metrics that can be integrated into existing monitoring, reporting, and alerting tools—so you can more effectively monitor model performance, reduce risk of model failure, and comply with internal and external regulations.
Algorithmia Insights continues our integration-first approach to MLOps by making critical model data and metrics available to existing management and operational tools. Users can export operational and inference metrics from every algorithm prediction and connect the metrics output to the tool of their choice.
Effectively manage your ML performance
Organizations need to monitor and report on ML performance for a variety of purposes:
- To report compliance with external and internal regulations
- To monitor model performance in order to improve business outcomes
- To reduce the risk of model failure, including identifying and correcting for model drift
However, many organizations today don’t have the ability to monitor model performance, and those that do are using a patchwork of disparate tools and manual processes, often without critical data required to satisfy stakeholder requirements. Without comprehensive AI/ML monitoring, organizations struggle with model drift, risk of failure, and inability to meet performance targets in response to shifts in environment and customer behavior.
Algorithmia Insights solves this problem by providing a centralized location to collect model performance data and connect it with tools that are already in use across the organization. Organizations are already using a variety of monitoring, reporting, and visualization systems for business and operational management, including:
- Operational dashboards for IT application and service monitoring
- Compliance tools for audit, explainability, and traceability of historical data
- BI tools for analysis
With Algorithmia Insights, companies can now emit a rich metrics payload from every prediction and integrate it into nearly any external logging, monitoring, or data visualization system. Identify model drift, improve model performance, and manage audit and compliance needs, all in the same systems you’re using to manage your existing business applications.
How it works
Algorithmia Insights offers a way for customers to port their model’s performance metrics to an external system where they can then save those model predictions along with metadata that can be used for traceability.
An Algorithmia Insights metrics pipeline consists of a structured payload holding both operational and model inference data that is streamed into Apache Kafka. It can include data such as:
- Model operational metrics
- Model inference metrics
- Data inputs
- Prediction outputs
- Model metadata
You can then connect that data to a wide range of popular monitoring, reporting, and alerting tools.
Get started with Algorithmia Insights today
Algorithmia Insights is available today and automatically included in the subscription for Algorithmia Enterprise, both the Dedicated and Advanced editions. Whether you’re an existing customer or are evaluating options for an MLOps platform, join us for a webinar on November 17 to learn what Algorithmia Insights can unlock for your organization. We’ll demonstrate Algorithmia Insights in action, discuss its variety of use cases, and show how to get started today.