Skip to content

Retrieval - Building Blocks

Welcome to the Retrieval Building Blocks documentation. These accelerators enable AI applications to access, query, and interact with data through various interfaces and storage mechanisms.

Overview

Retrieval capabilities provide the "data access layer" for AI applications, enabling semantic search, NoSQL storage, and efficient federated data retrieval across multiple sources.

Available Building Blocks

RAG (Retrieval-Augmented Generation)

Complete RAG pipeline with document ingestion, embedding generation, vector storage, and semantic search capabilities.

Key Features:

  • Document ingestion from IBM Cloud Object Storage
  • Embedding generation with IBM Watsonx.ai
  • Vector storage (Milvus or OpenSearch)
  • Semantic, keyword, and hybrid search
  • MCP server integration for AI assistants
  • Bob modes for RAG development guidance

Components:

  • RAG Accelerator: Complete pipeline with FastAPI REST API
  • RAG Ingestion MCP Server: Document ingestion for AI assistants
  • RAG Retrieval MCP Server: Semantic and keyword search
  • Bob Modes: AI assistant modes for RAG development

Vector ingestion, embedding, and retrieval for semantic similarity search in GenAI pipelines.

Key Features:

  • Document parsing and extraction
  • Multiple embedding strategies (dense, hybrid, dual)
  • Flexible chunking strategies
  • REST API with authentication

Supported Databases:

No SQL Database

Large-scale NoSQL storage with Cassandra compatibility and optional vector capabilities for AI and application workloads.

Key Features:

  • Apache Cassandra-based serverless database
  • Vector collections for AI applications
  • Data API and CQL support
  • Scalable and highly available

Zero Copy

Federated analytics without copying data. Query data across distributed sources with open lakehouse architecture.

Key Benefits:

  • Cost Savings: No redundant storage costs
  • Faster Insights: Avoids ETL delays
  • Single Source of Truth: Reduces data inconsistencies
  • Flexibility: Multiple engines access the same data
  • Governance: Centralized access control

IBM Products:

  • IBM watsonx.data
  • IBM Cloud Object Storage (COS)
  • IBM Db2 Database
  • Presto Query Engine

Use Cases

Common Retrieval Scenarios

  • RAG Systems: Build complete Retrieval-Augmented Generation pipelines
  • Question Answering: Intelligent Q&A over document collections
  • Semantic Search: Find documents based on meaning, not just keywords
  • Hybrid Search: Combine semantic understanding with keyword precision
  • Knowledge Management: Create searchable knowledge bases from unstructured data
  • AI Assistant Integration: Add RAG capabilities via MCP servers
  • Multi-Cloud Analytics: Query data across AWS, IBM Cloud, and on-premises
  • Real-Time Insights: Access live data without ETL delays
  • NoSQL Storage: Scalable storage for AI application data

Resources

Support

For issues or questions, please refer to the GitHub repository or contact IBM support.