RAG System Local Deployment Design

# RAG System Local Deployment Design

You are a senior RAG architect. Design a production-quality local RAG system for the specified knowledge domain.

## System Requirements

- Knowledge domain: [DOMAIN, e.g., "legal case library", "internal engineering wiki", "product documentation"]
- Document count: [APPROXIMATE NUMBER]
- Average document length: [PAGES OR WORDS]
- Query volume: [QUERIES PER DAY]
- Target response latency: [ACCEPTABLE WAIT TIME]
- Hardware budget: [HARDWARE SPECS OR BUDGET]

## Architecture

### Chunking Strategy
Recommend chunk size, overlap, and splitting strategy (sentence, paragraph, semantic, or hybrid) for [DOMAIN]. Justify based on expected query type.

### Embedding Pipeline
Local embedding model selection, batch size for indexing, and estimated indexing time for [DOCUMENT COUNT] documents.

### Vector Store Selection
Compare ChromaDB, Qdrant, and pgvector for this workload. Select one and explain the production trade-offs.

### Retrieval Layer
Design the retrieval pipeline: number of retrieved chunks, re-ranking step, context window assembly, and metadata filtering strategy.

### Generation Layer
Local LLM selection, system prompt template, and hallucination mitigation techniques for [DOMAIN].

### Observability
How to log retrieval quality, track answer drift over time, and alert when the system misses questions it should answer.