RAG Pipeline

RAG Pipeline An architecture diagram generated by Archify. User · Chat / API · Architecture component User Chat / API Guardrail · Input filter · Query Runtime · PII + toxicity Guardrail Input filter PII + toxicity Orchestrator · LangGraph agent · Query Runtime Orchestrator LangGraph agent Semantic Cache · Redis :6379 · Query Runtime Semantic Cache Redis :6379 Retriever · Hybrid search · Query Runtime Retriever Hybrid search Vector Store · Milvus · Architecture component · HNSW index Vector Store Milvus HNSW index Reranker · Cross-encoder · Query Runtime Reranker Cross-encoder LLM · Claude API · Query Runtime LLM Claude API Documents · PDF / Web / DB · Architecture component Documents PDF / Web / DB Loader · Parse + extract · Ingestion Pipeline Loader Parse + extract Chunker · Semantic split · Ingestion Pipeline Chunker Semantic split Embedding · text-embedding-3 · Ingestion Pipeline Embedding text-embedding-3 question validated query semantic match retrieve ANN search top-k docs context grounded answer response raw docs parsed text chunks vectors Query Runtime Ingestion Pipeline Legend Backend Database Cloud Security External

Retrieval

  • • Hybrid search: dense vectors + sparse BM25
  • • Cross-encoder reranker for precision
  • • Top-k context window management

Generation

  • • Claude API for grounded generation
  • • Semantic cache reduces repeat LLM calls
  • • Streaming response with citations

Safety

  • • Input guardrail filters PII and toxic content
  • • Output grounding reduces hallucination risk
  • • Document-level access control in retriever