search-pipeline

name	Search Pipeline
description	RAG Modulo search architecture including pipeline resolution, Chain of Thought reasoning, MCP gateway, and agentic search patterns. Trigger when modifying search flow, CoT, or MCP integration.
allowed-tools	Read, Grep, Glob, Write, Edit, Bash

Search Pipeline Architecture

Core Flow

1. Request arrives at router/search_router.py -> SearchService.search()
2. Pipeline resolved automatically: _resolve_user_default_pipeline()
3. Complex question detection: _should_use_chain_of_thought()
4. If CoT: ChainOfThoughtService orchestrates decomposition + synthesis
5. Vector search via configured provider (Milvus default)
6. Response assembly with source attribution

Key Files

• Entry: backend/rag_solution/services/search_service.py
• CoT: backend/rag_solution/services/chain_of_thought_service.py
• Schema: backend/rag_solution/schemas/search_schema.py
• MCP Gateway: backend/rag_solution/services/mcp_gateway_client.py
• MCP Server: backend/mcp_server/server.py

CoT Hardening

• Structured output: XML <thinking> / <answer> tags
• Multi-layer parsing: XML -> JSON -> markers -> regex -> full response
• Quality scoring: 0.0-1.0 confidence with artifact detection
• Retry logic: up to 3 attempts, quality threshold 0.6
• Docs: docs/features/chain-of-thought-hardening.md