RAG Evaluation and Retrieval Blueprint Template

When to use this

Architecture document template for production Retrieval-Augmented Generation (RAG) systems.

The template

# RAG Pipeline Architecture Spec

## Pipeline Overview

1. **Ingestion & Chunking**: Recursive character chunking (500 token chunk size, 50 token overlap).
2. **Embedding Model**: OpenAI `text-embedding-3-small` or Cohere Embed v3.
3. **Vector Database**: Qdrant / Pgvector / Supabase with HNSW index.
4. **Re-Ranking**: Cohere Rerank v3 filtering top 20 candidates down to top 5.
5. **Evaluation Suite**: Ragas framework measuring Context Precision, Context Recall, and Faithfulness scores.

## Component Flow

```mermaid
graph TD
    A[User Query] --> B[Embedding Generator]
    B --> C[Vector DB HNSW Search]
    C --> D[Top 20 Chunks]
    D --> E[Cohere Re-Ranker]
    E --> F[Top 5 Relevant Chunks]
    F --> G[LLM Context Generator]
    G --> H[Response Output]
```

How to use

Save to docs/architecture/rag-pipeline.md.