Agent Skills: Design RAG Architecture

Design RAG Architecture

Design a Retrieval-Augmented Generation system for a given use case.

Arguments

$ARGUMENTS - The RAG use case to design for (e.g., "customer support chatbot", "documentation Q&A", "legal document search", "code assistant")

Workflow

1. Clarify requirements by understanding:

   • What type of questions will be asked?
   • What is the document corpus size and type?
   • What is the required accuracy/faithfulness?
   • What is the latency budget?
   • Are there multi-turn conversation requirements?

2. Load relevant skills based on the use case:

   • RAG patterns → rag-architecture
   • Vector store selection → vector-databases
   • LLM serving → llm-serving-patterns
   • Inference optimization → ml-inference-optimization

3. Spawn the rag-architect agent for comprehensive design:

   • Use Task tool with subagent_type="rag-architect"
   • Provide full use case context and requirements
   • Request end-to-end RAG architecture

4. Design the ingestion pipeline:

   • Document extraction (PDF, HTML, code)
   • Chunking strategy selection
   • Embedding model selection
   • Vector database configuration
   • Metadata extraction and indexing

5. Design the retrieval pipeline:

   • Query processing (expansion, HyDE)
   • Retrieval strategy (dense, sparse, hybrid)
   • Reranking approach
   • Context assembly
   • Prompt engineering

6. Address quality and scale:

   • Retrieval accuracy (recall@k, MRR)
   • Answer faithfulness (grounding)
   • Latency budget allocation
   • Cost optimization
   • Scaling strategy

Example Usage

/sd:rag-design customer support chatbot with 10K FAQ documents
/sd:rag-design internal documentation Q&A for engineering team
/sd:rag-design legal document search for contract review
/sd:rag-design code assistant for enterprise codebase
/sd:rag-design research paper Q&A with 100K papers
/sd:rag-design product catalog search with structured data
/sd:rag-design multi-lingual knowledge base

Use Case Categories

| Category | Key Considerations | | -------- | ------------------ | | Customer Support | FAQ coverage, escalation, tone consistency | | Documentation | Technical accuracy, code examples, versioning | | Legal/Compliance | Citation accuracy, audit trails, access control | | Code Assistance | AST-aware chunking, context relevance, IDE integration | | Research/Academic | Multi-document reasoning, citation, long-form answers | | E-commerce | Product attributes, inventory awareness, personalization |

RAG Pattern Selection Guide

| Complexity | Pattern | When to Use | | ---------- | ------- | ----------- | | Low | Basic RAG | Simple Q&A, small corpus | | Medium | RAG + Reranking | Higher accuracy needed | | Medium | Hybrid Search | Mixed keyword + semantic queries | | High | Query-Transformed | Vague or complex queries | | High | Agentic RAG | Multi-hop reasoning, tool use |

Output

A comprehensive RAG system architecture including:

• Ingestion pipeline (documents → vectors)
• Retrieval pipeline (query → context)
• Technology stack (embedding model, vector DB, LLM)
• Quality targets (recall, faithfulness, latency)
• Trade-offs and alternatives
• Cost estimate (per-query and monthly)