DataRobot Fall 2023

Closing the Generative AI Confidence Gap

DataRobot’s newest release gives you the confidence to achieve real-world value with generative AI, enabling you to rapidly build with optionality, govern with full transparency, and operate with correctness and control.

Operate with Correctness and Control

Continuously improve models with comprehensive insights and guardrails while maintaining data security at scale.

Unified Console: Observe all models in one unified location before production – generative and predictive, regardless of their origin.

GenAI Guard Models and Human Feedback Loop: Combine generative AI with predictive models to evaluate attributes like toxicity, sentiment, personal information leaks, prompt ingestion, and correctness for models in production. Continuously improve your models through user feedback on the quality of their generated responses.

Out of the Box and Custom Metrics: Evaluate generative and predictive models using pre-defined metrics such as text drift and service health, and configure use case-specific metrics to measure performance against desired business goals. Receive real-time notifications if metrics exceed designated thresholds to quickly intervene.

Model Monitoring
Model Monitoring

Govern with Full Transparency

Govern all AI from one unified location – generative and predictive models, built on or off-platform.

Unified Model Registry: Test and evaluate both generative and predictive models in one unified location before production, regardless of their origin.

GenAI Cost Insights Metrics: Monitor GenAI costs in real-time at the deployment level with customizable metrics for informed cost-performance decisions; receive alerts when costs surpass specified thresholds.

Enterprise-Grade LLM Deployments: Register each deployment as one flow and monitor the complete GenAI ‘recipe,’ set workflow approvals and permissions for production changes, and enable effortless version control.

Custom Job Insights: Create groups of tests with custom environments and dependencies for model approval, scenario checks and personalized explainability insights.

Batch-Based Monitoring: Track and manage model monitoring statistics, such as data drift and accuracy, by specific batch jobs.

Global/Public Models in Registry: Access public models added by DataRobot or your organization’s admin, such as a Toxicity Classifier from Hugging Face, without requiring each user to register or explicitly share them.

Notification Policies: Configure and get model performance and health notifications instantaneously.

Custom Generative AI Model Metrics
Generative AI Custom Performance Metrics

Rapidly Build with Optionality

Rapidly deploy new GenAI use cases using an intuitive interface to experiment with the underlying components of your choice.

Multi-Provider LLM Playground: First-of-its-kind visual comparison interface with out-of-the-box access to external LLM services, including Google PaLM, Azure OpenAI, and AWS BedRock, or the ability to bring your own – to easily compare and experiment with different GenAI ‘recipes’ with any combination of foundation models, vector databases, and prompting strategies.

Model Comparison: Easily and visually compare insights and lineage across model experiments within a single use case.

Network Access for Custom Tasks: Customize your model blueprint pipelines to integrate external models or services through APIs, including your own LLMs and external embedding.

Vector Database Builder: Extend LLMs with your company data to protect data privacy and ensure actionable responses for your organization; easily track the lineage of any vector database created within our Notebook or the UI.

Enterprise Messaging App Integrations: Seamlessly integrate with the messaging apps your organization uses, like Slack and Microsoft Teams, to connect with end users and ensure widespread adoption.

Notebook Scheduling: Optimize your workflow by scheduling or triggering your notebook, all while maintaining detailed execution tracking automatically.

Notebook Custom Environment Integration: Define and reuse custom dependencies and environments within notebook sessions and custom models.

GPUs (Public Preview): Leverage automated GPU utilization when working with unstructured data, whether you’re coding or using the GUI, all while maintaining safeguards.

GenAI Accelerators: Use our pre-built AI Accelerators templates to quickly build GenAI specific industry and department use cases, or to enable LLMOPs and observability of GenAI solutions built with Google PaLM, Azure OpenAI, AWS BedRock, and more.  

Data Integration Enhancements: Seamlessly connect and wrangle data in Databricks or materialize data in BigQuery.

Advanced Modeling Settings in Workbench: Explore advanced modeling settings within the new Workbench experience.

Time Series in Workbench: Utilize our time series modeling capabilities within the new Workbench experience.

Generative AI Playground
Multi-Provider LLM Playground

Go the DataRobot Documentation Release Center for more information.