How aws-samples/rag-workshop-amazon-bedrock-knowledge-bases works
This repository is a collection of Amazon Bedrock RAG examples, notebooks, and supporting Python utilities—not a standalone application. Its main workflow is to prepare and ingest documents into a knowledge base, retrieve relevant context, and generate answers; companion examples explore evaluation and accuracy improvements. Separate branches cover responsible AI, multimodal, structured, and graph RAG. The tree and sampled utilities support these capabilities, but notebook contents were not sampled, so detailed notebook-to-utility wiring is not asserted.
Main entry points: 01_create_ingest_documents_test_kb_multi_ds.ipynb Ingest documents · synthetic_dataset Sample media · 02_managed_retreiveandgenerate_and_streamapi.ipynb Managed retrieval · 03_customized_rag_retreive_api_hybrid_search_langchain.ipynb Hybrid search
Components 25
RAG foundations
- Ingest documents
01-rag-concepts/01_create_ingest_documents_test_kb_multi_ds.ipynb - Managed retrieval
01-rag-concepts/02_managed_retreiveandgenerate_and_streamapi.ipynb - Hybrid search
01-rag-concepts/03_customized_rag_retreive_api_hybrid_search_langchain.ipynb - RAG evaluation
01-rag-concepts/04_customized_rag_retreive_api_langchain_evaluation_ragas.ipynb - Document ingestion
01-rag-concepts/05_document_level_kb_ingestion.ipynb
Retrieval quality
- Chunking options
02-optimizing-accuracy-retrieved-results/advanced_chunking_options.ipynb - Metadata filters
02-optimizing-accuracy-retrieved-results/metadata_filtering.ipynb - Generated metadata
02-optimizing-accuracy-retrieved-results/autogenerated_metadata_filters.ipynb - CSV metadata
02-optimizing-accuracy-retrieved-results/csv_metadata_customization.ipynb - Query reformulation
02-optimizing-accuracy-retrieved-results/query_reformulation.ipynb - Result reranking
02-optimizing-accuracy-retrieved-results/re_ranking_using_kb.ipynb
RAG extensions
- Context grounding
03-responsible-ai/contextual_grounding.ipynb - Audio/video prep
04-multi-modal-rag/audio-video-rag-using-kb/01_data_prep_using_bda.ipynb - Audio/video RAG
04-multi-modal-rag/audio-video-rag-using-kb/02_audio_video_rag_using_kb.ipynb - Text/image RAG
04-multi-modal-rag/text-images-rag-using-kb/text_images_rag_using_kb.ipynb - Structured RAG
05-structured-rag/structured_rag_using_kb.ipynb - Graph RAG
06-graph-rag/01-graph-rag-using-kb.ipynb
Shared utilities
- Knowledge base setup
utils/knowledge_base.py - KB operations
utils/knowledge_base_operators.py - Managed KB helper
utils/managed_knowledge_base.py - Structured KB helper
utils/structured_knowledge_base.py - Evaluation helper
utils/evaluation.py
Other components
- Notebook user
- AWS Bedrock
- Sample media
synthetic_dataset
Connections 27
- Notebook user → Ingest documents: explores ingestion
- Notebook user → Managed retrieval: explores retrieval
- Notebook user → Hybrid search: explores search
- Notebook user → RAG evaluation: runs evaluation
- Sample media → Audio/video prep: provides media
- Sample media → Ingest documents: provides documents
- Ingest documents → AWS Bedrock: creates knowledge base
- Managed retrieval → AWS Bedrock: retrieves and generates
- Hybrid search → AWS Bedrock: retrieves context
- RAG evaluation → AWS Bedrock: evaluates answers
- Document ingestion → AWS Bedrock: ingests documents
- Chunking options → AWS Bedrock: configures chunking
- Metadata filters → AWS Bedrock: filters retrieval
- Generated metadata → AWS Bedrock: generates metadata
- CSV metadata → AWS Bedrock: customizes metadata
- Query reformulation → AWS Bedrock: reformulates queries
- Result reranking → AWS Bedrock: reranks results
- Context grounding → AWS Bedrock: checks grounding
- Audio/video prep → AWS Bedrock: prepares media
- Audio/video RAG → AWS Bedrock: retrieves media context
- Text/image RAG → AWS Bedrock: retrieves image context
- Structured RAG → AWS Bedrock: queries structured data
- Graph RAG → AWS Bedrock: uses graph retrieval
- Knowledge base setup → AWS Bedrock: creates KB resources
- Managed KB helper → AWS Bedrock: creates managed KB
- Structured KB helper → AWS Bedrock: creates structured KB
- Evaluation helper → AWS Bedrock: retrieves and evaluates