Caching Layer Design for RAG

Design a comprehensive caching layer for a RAG system.

## RAG Pipeline
{{rag_pipeline}}

## Traffic Patterns
{{traffic_patterns}}

## Cache Infrastructure
{{cache_infrastructure}}

Design multi-level caching:

**Level 1: Query Cache**
- Exact query match
- Semantic similarity cache
- TTL and invalidation strategy

**Level 2: Embedding Cache**
- Pre-computed query embeddings
- Document embeddings
- Cache key design

**Level 3: Retrieval Result Cache**
- Query-to-results mapping
- Partial result caching
- Freshness vs latency trade-off

**Level 4: LLM Response Cache**
- Prompt-response pairs
- Context-aware caching
- Output variability handling

**Cache Orchestration**
- Cache lookup order
- Cache population strategy
- Metrics and monitoring

Provide implementation architecture and expected hit rates.