Agentic Design

Patterns
📚

Naive RAG(NRAG)

Foundational "Retrieve-Read" framework following traditional indexing, retrieval, and generation process

Complexity: lowKnowledge Retrieval (RAG)

🎯 30-Second Overview

Pattern: Simple retrieve-then-read pipeline: query → retrieve documents → concatenate → generate

Why: Provides external knowledge access with minimal complexity - foundational approach established by Lewis et al. (2020)

Key Insight: Direct concatenation of top-k retrieved documents to query prompt - no optimization or post-processing

⚡ Quick Implementation

1Index:Create vector embeddings of knowledge base
2Retrieve:Find top-k relevant documents via similarity
3Concatenate:Append retrieved docs to query prompt
4Generate:Submit combined prompt to LLM
5Return:Output generated response directly
Example: query → vector_search → concat_context → llm_generate → response

📋 Do's & Don'ts

✅Use semantic embeddings (e.g., sentence-transformers)
✅Implement basic chunking strategy (fixed size 512-1024 tokens)
✅Set clear top-k retrieval limits (typically 3-5 documents)
✅Include document metadata and source attribution
✅Use FAISS or similar for efficient vector search
❌Skip query preprocessing or normalization
❌Retrieve too many documents (causes context pollution)
❌Ignore relevance scoring thresholds
❌Concatenate without clear document boundaries
❌Expect sophisticated reasoning from basic approach

🚦 When to Use

Use When

  • • Simple Q&A over documents
  • • Proof-of-concept RAG systems
  • • Limited technical complexity allowed
  • • Small to medium knowledge bases
  • • Straightforward factual queries

Avoid When

  • • Complex multi-hop reasoning required
  • • High-accuracy critical applications
  • • Large-scale production systems
  • • Noisy or contradictory knowledge bases
  • • Real-time performance requirements

📊 Key Metrics

Retrieval Accuracy
Relevant docs in top-k (Recall@k)
Answer Quality
BLEU/ROUGE scores vs ground truth
Response Time
End-to-end latency (retrieval + generation)
Context Utilization
% of retrieved context used in response
Hallucination Rate
% responses with unsupported claims
Source Attribution
% responses with correct source citations

💡 Top Use Cases

Document Q&A: Simple factual questions over company documents or knowledge bases
FAQ Systems: Automated responses to frequently asked questions using existing documentation
Research Assistance: Basic information retrieval from academic papers or research collections
Customer Support: Level 1 support queries answerable from documentation and manuals
Educational Tools: Simple question-answering over textbooks and educational materials

References & Further Reading

Deepen your understanding with these curated resources

Foundational Papers

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al., 2020)

Dense Passage Retrieval for Open-Domain Question Answering (Karpukhin et al., 2020)

RAG vs FiD: Comparing Retrieval-Augmented Generation Models (Izacard & Grave, 2021)

Leveraging Passage Retrieval with Generative Models (Izacard et al., 2022)

Comprehensive Surveys

Retrieval-Augmented Generation for Large Language Models: A Survey (Gao et al., 2023)

A Comprehensive Survey of RAG: Evolution and Future Directions (Gupta et al., 2024)

Retrieval-Augmented Generation for AI-Generated Content: A Survey (Li et al., 2024)

RAG and RAU: A Survey on Retrieval-Augmented Language Model (Zhao et al., 2024)

Implementation Resources

LangChain RAG Tutorial - Basic Implementation

Hugging Face RAG Documentation and Examples

OpenAI RAG Implementation Guide

LlamaIndex Basic RAG Pipeline Tutorial

Vector Database Solutions

FAISS - Facebook AI Similarity Search Library

Pinecone Vector Database Documentation

Weaviate Vector Search Engine

Chroma - Open-source Embedding Database

Embedding Models & Evaluation

Sentence-BERT: Sentence Embeddings using Siamese BERT (Reimers & Gurevych, 2019)

Text and Code Embeddings by Contrastive Pre-Training (OpenAI, 2022)

MTEB: Massive Text Embedding Benchmark (Muennighoff et al., 2022)

BGE: BAAI General Embedding Model Documentation

Evaluation Frameworks

RAGAS: Evaluation Framework for RAG Applications

TruLens: Evaluation and Tracking for LLM Applications

LangSmith: LLM Application Testing and Monitoring

DeepEval: Unit Testing for LLM Applications

Contribute to this collection

Know a great resource? Submit a pull request to add it.

Contribute

Patterns

closed

Loading...

Built by Kortexya