Our product team has been busy this week and we’re excited to share some of the updates we’ve shipped. We thank you for your continued support and collaboration, since it directly helps us continue to ship great products for our customers.

Product Updates

Role Based Access Control (RBAC) for Models

With this feature we aim to create a robust way to manage resources in large organizations to ensure that model access is easily assigned and administered with fine-grained differential permissions. Default Roles: [Platform Administrator, Application Administrator, Developer]

Note: Exact policies and roles will vary per organization and are configurable. These are the default roles Radiant ships with. Reach out to your Radiant contact for more details.

• Platform administrators can set policies for applications based on individuals or groups.

• Platform Administrators can set budgets for individuals, groups, applications, and/or models.

• Application Administrators can set policies for applications based on individuals or groups.

Dynamic Routing Policies

Administrators can now set policies to control flow to models based on attributes like request payload, metadata, and more. The policies can be customized for each enterprise, Radiant ships policies out of the box:

Policies to control flow to models based on the content being set. Some example use cases:

• Blocking certain types of content (keyword, regular expression, and more) from being sent to any models

• Re-routing certain types of content (keyword, regular expression, and more) to specific models

Policies to control flow to models based on request metadata including user id, application id, or more. Some examples of how this can be used:

• For the same application, ensure that EU-based users are routed to a model that is GDPR-compliant, but pick the best available model for US-based users

• Route request from teams or groups working with sensitive (PII, HIPAA, etc) data to secure models, but pick the cheapest available model for other groups

Multi-Model and Prompt Evaluations Suite

Users can test and evaluate AI applications during development and production via Radiant. Radiant’s evaluation suite takes advantage of our existing observability and audit frameworks to enable:

• Testing and comparing models and prompts when you are setting routing logic for multi-model applications

• Detecting regressions in new models you are using, and automatically routing back to the last “known good” configuration

• Analyzing model prompts and responses in bulk and flagging anomalies for human review

Other Updates

• We co-wrote a blog post with Airbyte on scaling embedding pipelines across models

About Radiant

Radiant is the Enterprise AI platform to take you from idea to production deployment. From security and governance to scaling and resiliency, we make it simple to make AI a critical part of your business. 

Try out a demo here, sign up here to get your own instance, or reach out to our founders directly at [email protected].

We’re also hiring. If you know of someone great that is interested in helping every company build AI into their products and operations, we'd love an introduction.