domains / data / rag-implementation / rag-implementation

rag-implementation

Skill card

combined

The three copies overlap on the core RAG pipeline (embeddings, vector store, chunking, retrieval, reranking, evaluation), but each contains material steps or constraints the others lack. rmyndharis/antigravity-skills provides a concise 4-step procedure plus safety constraints (redaction, access controls, source-document exposure) and a LangChain quick start. sickn33/agentic-awesome-skills adds a phased workflow with explicit phases for LLM integration, caching, and evaluation, plus quality gates and limitations. LeoYeAI/openclaw-master-skills provides a LangGraph quick start, a 2026 embedding model table, and a different vector database option (pgvector). Crowning one copy would hide the other two's unique steps and constraints.

Repositories

Other copies

Combined skill

---
name: rag-implementation
description: Build Retrieval-Augmented Generation (RAG) systems for LLM applications with vector databases and semantic search. Use when implementing knowledge-grounded AI, building document Q&A systems, or integrating LLMs with external knowledge bases.
---

# RAG Implementation

Master Retrieval-Augmented Generation (RAG) to build LLM applications that provide accurate, grounded responses using external knowledge sources.

## Use this skill when

- Building Q&A systems over proprietary documents
- Creating chatbots with current, factual information
- Implementing semantic search with natural language queries
- Reducing hallucinations with grounded responses
- Enabling LLMs to access domain-specific knowledge
- Building documentation assistants
- Creating research tools with source citation

## Do not use this skill when

- You only need purely generative writing without retrieval
- The dataset is too small to justify embeddings
- You cannot store or process the source data safely

## Instructions

1. Define the corpus, update cadence, and evaluation targets.
2. Choose embedding models and vector store based on scale.
3. Build ingestion, chunking, and retrieval with reranking.
4. Integrate the LLM: select provider, design prompt template, implement context injection, and add citation handling.
5. Implement caching: response caching, embedding cache, TTL, cache invalidation, and hit-rate monitoring.
6. Evaluate with grounded QA metrics and monitor drift.

## Safety

- Redact sensitive data and enforce access controls.
- Avoid exposing source documents in responses when restricted.

## Core Components

### 1. Vector Databases
**Purpose**: Store and retrieve document embeddings efficiently

**Options:**
- **Pinecone**: Managed, scalable, fast queries
- **Weaviate**: Open-source, hybrid search
- **Milvus**: High performance, on-premise
- **Chroma**: Lightweight, easy to use
- **Qdrant**: Fast, filtered search
- **FAISS**: Meta's library, local deployment
- **pgvector**: PostgreSQL extension, SQL integration

### 2. Embeddings
**Purpose**: Convert text to numerical vectors for similarity search

**Models:**
- **text-embedding-ada-002** (OpenAI): General purpose, 1536 dims
- **all-MiniLM-L6-v2** (Sentence Transformers): Fast, lightweight
- **e5-large-v2**: High quality, multilingual
- **Instructor**: Task-specific instructions
- **bge-large-en-v1.5**: SOTA performance
- **voyage-3-large**: 1024 dims, recommended for Claude apps
- **voyage-code-3**: 1024 dims, code search
- **text-embedding-3-large**: 3072 dims, high accuracy
- **text-embedding-3-small**: 1536 dims, cost-effective
- **multilingual-e5-large**: 1024 dims, multi-language support

### 3. Retrieval Strategies
**Approaches:**
- **Dense Retrieval**: Semantic similarity via embeddings
- **Sparse Retrieval**: Keyword matching (BM25, TF-IDF)
- **Hybrid Search**: Combine dense + sparse
- **Multi-Query**: Generate multiple query variations
- **HyDE**: Generate hypothetical documents

### 4. Reranking
**Purpose**: Improve retrieval quality by reordering results

**Methods:**
- **Cross-Encoders**: BERT-based reranking
- **Cohere Rerank**: API-based reranking
- **Maximal Marginal Relevance (MMR)**: Diversity + relevance
- **LLM-based**: Use LLM to score relevance

## Quick Start

Use either a LangChain RetrievalQA chain or a LangGraph StateGraph depending on your framework preference. Both follow the same pipeline: load documents, split into chunks, create embeddings and vector store, build a retriever, and generate a grounded answer.

## Advanced RAG Patterns

### Pattern 1: Hybrid Search
Combine BM25 sparse retrieval with dense embedding retrieval using weighted fusion or Reciprocal Rank Fusion.

### Pattern 2: Multi-Query Retrieval
Generate multiple query variations from a single user query to improve recall.

### Pattern 3: Contextual Compression
Use an LLM-based compressor to extract only relevant portions of retrieved documents before passing them to the generator.

### Pattern 4: Parent Document Retriever
Retrieve small child chunks for matching, then return the larger parent chunk for context.

## Document Chunking Strategies

Use recursive character splitting with configurable chunk size and overlap. Tune chunk size and overlap based on document structure and retrieval quality.

## Quality Gates

- [ ] Embedding model selected
- [ ] Vector DB configured
- [ ] Chunking implemented
- [ ] Retrieval working
- [ ] LLM integrated
- [ ] Caching configured
- [ ] Evaluation passing

## Limitations

- Use this skill only when the task clearly matches the scope described above.
- Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
- Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.