DataRobot Vector Database

Category: Managed Vector Databases
Brand: DataRobot
Website: https://www.datarobot.com/platform/generative-ai/

DataRobot Vector Database is a managed vector store within the DataRobot AI Platform, designed to power production-grade vector search use cases such as RAG (Retrieval-Augmented Generation) and semantic search. It supports NVIDIA NIM embeddings and both built-in and bring-your-own embedding models, and can be combined with other external vector databases as part of broader generative AI workflows.

Features

Managed Vector Store for AI Workloads

• Managed vector database capability integrated into the DataRobot AI Platform.
• Designed for AI workloads including:
  • Retrieval-Augmented Generation (RAG)
  • Semantic search
  • Generative AI apps that "talk to" tabular data and documents
• Production-oriented vector search for enterprise use cases.

Lifecycle Management

• Create, register, deploy, and update vector databases directly within the platform.
• Treat vector databases as first-class assets within generative AI workflows.

Embedding Model Support

• Integration with NVIDIA NIM embeddings.
• Support for both:
  • Built-in embeddings managed by DataRobot.
  • Bring-your-own embedding models.
• Can be used alongside the platform’s LLM gateway to mix and match over 70 generative AI and embedding models (no separate credential management required at the application level).

RAG and Pipeline Integration

• Built for building and running advanced RAG systems in production.
• Can be combined in pipelines with:
  • LLMs and SLMs (small language models)
  • Retrievers
  • Prompts
• Allows creating a single unified endpoint that encapsulates LLM, vector DB, retriever, and prompt logic, callable from any application.

Interoperability with External Vector Databases

• Ability to bring in external vector databases (e.g., Pinecone, Elasticsearch) to build or augment RAG workflows.
• Supports workflows that span both DataRobot’s managed vector database and third-party vector stores.

Experimentation and Tuning

• Rapid experimentation across the generative AI workflow:
  • Test and compare different embedding models and configurations.
  • Experiment with multiple chunking strategies for documents.
  • Adjust context features (e.g., context window, retrieval settings) to optimize vector search and RAG quality.
• Prompt and model experimentation tooling to select the best-performing configuration in conjunction with the vector database.

Evaluation and Benchmarking (Vector-Backed Workflows)

• Evaluate end-to-end workflows (including those backed by vector databases) on:
  • Faithfulness
  • Correctness
  • Latency
  • Cost
• Ability to synthetically generate large volumes of question–answer pairs across existing vector databases to:
  • Build or extend ground truth datasets.
  • Benchmark and compare different model + vector DB configurations.

Cloud & Infrastructure Integration

• Can participate in workflows that access GPUs across multiple clouds for inference and evaluation.
• Designed for latency, cost, and availability optimization in enterprise generative AI workloads.

Pricing

No specific pricing or plan details are provided in the available content. Users are directed to DataRobot’s main platform site and trial page for access and commercial information.