build-rag

Build or modify RAG pipeline: $ARGUMENTS

Before Starting

1. Search OpenAI docs for retrieval patterns:
   • Use mcp__openai-docs__search_openai_docs with query "vector stores" for vector store management
   • Use mcp__openai-docs__search_openai_docs with query "file search tool" for file_search tool configuration
   • Use mcp__openai-docs__search_openai_docs with query "retrieval augmented generation" for RAG best practices
2. Read the existing RAG implementation in ai-lab-agent:
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/ai-lab/retrieval/router.ts — intent classification and corpus routing
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/ai-lab/retrieval/search.ts — full pipeline (search → filter → re-sort → cite)
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/ai-lab/retrieval/citations/parseSource.ts — file path → book/chapter/section
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/ai-lab/retrieval/citations/renderSources.ts — citation line + sources block
3. Read the vector store integration:
   • {{PROJECT_ROOT}}/{{AGENTS_DIR}}/shared/tools.ts — executeFileSearch() using OpenAI Vector Store Search API
   • {{PROJECT_ROOT}}/src/lib/writing-assistant/file-search.ts — org-specific vector store lookup

RAG Pipeline Architecture