/projects-overview.webp)
A single system for all your models
Documentation, code, and model inventories inevitably get out of sync when spread across systems. Regulations like SR 11-7, CCAR, CRD IV, and GDPR require a holistic approach to model risk management that spans development and usage. Domino tracks the provenance of a model from idea to impact, showing who worked on it, what they did, how they deployed it, and how it is used in production.
Quickly reproduce and validate work
Model validation teams spend the vast majority of their time gathering documents to piece together how a model was built and performs. Domino's Reproducibility Engine automatically tracks changes to code, data, tools, and packages through continual version control. These are captured in Durable Workspaces that allow data scientists and validators to instantly recreate the exact experiment environment used to create a model, even if tools and developers have changed in the interim. The discussion functionality centralizes conversations previously buried in thousands of emails.
Out-of-the-box support for Feast– the emerging, open-source standard for feature stores – increases the reproducibility and reusability of features. You have a single source of truth for calculating important features, driving reuse and consistency.
Efficiently track all aspects of model performance
Many organizations struggle to effectively monitor models in production and efficiently remediate issues to keep models at peak performance.
Domino's integrated model monitoring provides a “single pane of glass” for observing traffic, drift, and health trends for all production models with out-of-the-box and custom metrics. You will be automatically alerted when drift, divergence, and data quality checks exceed thresholds. When retraining is needed, it's easy to drill down to model features to modify, retrain and redeploy models quickly.
This lets you see all production assets generated within Domino in a single location, acting as a model inventory. See all usage metrics, versions, and error rates for all data science models that have entered production inside your organization.