The Evolution of RAG Patterns

A companion repository showing how retrieval-augmented generation evolves from a simple vector lookup into more capable retrieval systems.

Featured Article

Read the full long-form guide here:

• The Ultimate Guide to RAG Patterns

It covers every implemented pattern in this repo, groups them into practical categories, includes Mermaid diagrams, and links directly to the code for each pattern.

Setup

1. Create and activate a virtual environment.
2. Install dependencies with pip install -r requirements.txt.
3. Set your OpenAI API key in your shell.
4. Run the demos in order.

Windows PowerShell:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
$env:OPENAI_API_KEY="your-key"
python .\01_naive_rag\naive_rag.py

macOS / Linux:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
export OPENAI_API_KEY="your-key"
python 01_naive_rag/naive_rag.py

Demo Sequence

1. 01_naive_rag: basic vector retrieval plus answer synthesis.
2. 02_advanced_rag: query expansion with a multi-query retriever.
3. 03_multi_step_rag: decomposes a complex question into sub-questions.
4. 04_agentic_rag: lets an agent choose between internal retrieval and web search.
5. 05_hybrid_rag: combines dense retrieval with keyword search.
6. 06_reranked_rag: retrieves a broader candidate set, then reranks it.
7. 07_metadata_filtered_rag: applies structured filters before semantic retrieval.
8. 08_parent_document_rag: retrieves child chunks but returns full parent sections.
9. 09_contextual_compression_rag: compresses retrieved documents before answering.
10. 10_corrective_rag: retries retrieval after rewriting a weak query.
11. 11_graph_rag: traverses graph-shaped facts instead of only chunk similarity.
12. 12_structured_data_rag: augments text retrieval with structured table data.
13. 13_conversational_rag: rewrites follow-up questions using chat history.
14. 14_citation_grounded_rag: answers with explicit source references.
15. 15_adaptive_router_rag: routes each query to the most relevant retriever.
16. 16_multimodal_rag: retrieves from text extracted out of slides, images, and diagrams.
17. 17_fusion_rag: fuses rankings from multiple retrievers and query variants.
18. 18_multi_hop_rag: performs chained retrieval across multiple hops.
19. 19_pdf_rag: retrieves and cites information from realistic invoice and tax-form PDFs.
20. 20_image_ocr_rag: retrieves from OCR-extracted text tied to realistic invoice and receipt images.
21. 21_local_image_ocr_rag: performs OCR locally on real document images before retrieval.

Data Layout

• data/structured/equipment_catalog.csv: structured product table used by mixed retrieval demos.
• data/semi_structured/policies.json: metadata-rich policy records.
• data/semi_structured/company_graph.json: graph facts for graph-based retrieval.
• data/semi_structured/multimodal_records.json: extracted records from slides, images, diagrams, and tables.
• data/semi_structured/image_ocr_records.json: OCR output linked to the enterprise images.
• data/semi_structured/receipt-result.json: extracted receipt analysis result.
• data/unstructured/text/company_handbook.txt: core handbook used by the first demos.
• data/unstructured/text/workspace_guides.txt: extra policy snippets for retrieval quality experiments.
• data/unstructured/text/employee_playbook.txt: larger sections for parent-child retrieval.
• data/unstructured/pdfs/enterprise_invoice_sample.pdf: real-looking invoice PDF sample.
• data/unstructured/pdfs/irs_1040_sample.pdf: official-looking tax-form PDF sample.
• data/unstructured/images/sample_invoice.jpg: real-looking invoice image sample.
• data/unstructured/images/receipt.png: real-looking scanned receipt sample.
• data/unstructured/images/contract.png: real-looking contract image sample.