Building Intelligent Search with Amazon Bedrock and Amazon OpenSearch for hybrid RAG solutions
New framework combines semantic and text search for more accurate AI assistants that handle complex queries.
AWS has introduced a new architectural blueprint for building more intelligent, agentic AI assistants by combining its managed services. The solution leverages Amazon Bedrock's foundational models and its AgentCore framework for orchestration, alongside Strands Agents for task management, and uses Amazon OpenSearch as a unified search backend. The core innovation is a hybrid RAG (Retrieval-Augmented Generation) approach that fuses semantic search—which finds data based on conceptual meaning using vector embeddings—with traditional keyword or lexical search. This combination is designed to overcome the limitations of using either method alone.
Semantic search, powered by bi-encoder models that create numerical vector representations of text, excels at understanding user intent and finding conceptually similar information, even without exact keyword matches. However, it can struggle with precise, fact-based queries. The new hybrid model allows an AI agent to first use semantic search to understand the broad context of a request (e.g., 'luxury hotel') and then apply text-based filters within OpenSearch to pinpoint exact criteria (e.g., 'in Miami, Florida'). This enables the creation of assistants that can manage multi-step, complex tasks like hotel booking by dynamically querying databases and APIs with greater accuracy.
The implementation detailed by AWS showcases a practical use case: a hotel booking assistant. When a user asks for a 'luxury hotel with ocean views in Miami,' the agentic system can decompose the request. It uses semantic search to understand concepts like 'luxury' and 'ocean views,' while simultaneously applying structured filters for location. The assistant then makes real-time API calls to check room availability and rates, integrating this live data into a coherent, conversational response for the user. This moves beyond simple chatbots to create dynamic systems that retrieve and act on business-specific data.
- Hybrid RAG architecture combines semantic vector search with traditional text-based search in Amazon OpenSearch.
- Uses Amazon Bedrock AgentCore and Strands Agents to orchestrate multi-step tasks and real-time API/data calls.
- Solves complex query challenges, like finding a 'luxury hotel with ocean views,' by understanding intent and applying precise filters.
Why It Matters
Enables enterprises to build AI agents that handle nuanced, real-world business queries with much higher accuracy and reliability.