diff --git a/README.md b/README.md index 35057c6..6ac6f1d 100644 --- a/README.md +++ b/README.md @@ -21,6 +21,7 @@ While [FAST](https://github.com/awslabs/fullstack-solution-template-for-agentcor | [LLM Council](#llm-council) | An implementation of "Council of LLMs" pattern on AWS. Builds consensus among multiple diverse LLMs.| | [Dual Monitoring System](#dual-monitoring-system) | Dual-layer monitoring for agentic solutions using AgentCore Evaluations and AWS DevOps Agent | | [AgentCore AWS Specialist Agent](#agentcore-aws-specialist-agent) | AWS specialist chat agent with Gateway MCP tools, long-term memory, web search, Skills on Runtime, and a NAT-free VPC | +| [Agentic Knowledge Discovery](#agentic-knowledge-discovery) | Combines a Bedrock Knowledge Base (documents) with an Aurora PostgreSQL metadata store (structured search), with metadata-filtered retrieval, page-level citations, and both Strands and LangGraph patterns | @@ -82,6 +83,18 @@ While [FAST](https://github.com/awslabs/fullstack-solution-template-for-agentcor  +### [Agentic Knowledge Discovery](samples/agentic-knowledge-discovery/) + +**Description**: A full-stack agent that answers questions over a mix of unstructured documents and structured metadata. It plans its retrieval — querying a metadata database to find relevant documents, then running a filtered semantic search over their content — and cites the exact source pages. + +**Built on FAST**: v0.4.2 + +**Key Differences from FAST**: Adds a Bedrock Knowledge Base (OpenSearch Serverless, Amazon Nova multimodal embeddings, hierarchical chunking, Bedrock Data Automation parsing) and an Aurora PostgreSQL metadata store; exposes two Gateway retrieval tools (`doc_search` and `structured_search`) that share a `doc_id` and metadata keys for combined structured + unstructured retrieval; adds page-level citations and follow-up suggestions; ships both Strands and LangGraph patterns behind AG-UI with in-UI model selection (Claude and GPT via Bedrock Mantle). + +**Use Case**: Building assistants that must reason across a document corpus and its structured metadata together — discovering the right documents by attribute, then searching only those, with grounded page-level citations. + + + + + + + +
+ + + +