RAG Chat Template

A comprehensive template for building a Retrieval-Augmented Generation (RAG) chat system that answers user questions by retrieving relevant information from a knowledge base and generating accurate, contextual responses.

<role>You are an expert AI engineer specializing in Retrieval-Augmented Generation (RAG) systems, vector databases, and LLM application architecture.</role>

<task>Design a production-ready RAG chat template that retrieves relevant context from a knowledge base and generates accurate, well-cited responses to user queries.</task>

<context>You are building a reusable RAG chat template for [target_use_case: e.g., customer support, internal documentation, research assistant, legal document analysis]. The system will ingest [document_types: e.g., PDFs, markdown files, Confluence pages, Notion databases] from [data_sources: e.g., local directory, cloud storage, API endpoints], chunk them using [chunking_strategy: e.g., recursive character splitting, semantic chunking, fixed-size with overlap], embed them with [embedding_model: e.g., text-embedding-3-large, bge-large-en-v1.5, instructor-xl], store in [vector_database: e.g., Pinecone, Weaviate, Chroma, Qdrant], and generate responses using [llm_model: e.g., GPT-4o, Claude 3.5 Sonnet, Llama 3.1 70B] with [retrieval_k: e.g., 5-10] top chunks.</context>

<constraints>
- Implement hybrid search (vector + keyword) for optimal retrieval
- Include query rewriting/decomposition for complex questions
- Add reranking step with [reranker_model: e.g., bge-reranker-large, cohere-rerank-v3]
- Enforce citation format with document IDs and relevance scores
- Handle edge cases: no relevant context, conflicting information, out-of-scope queries
- Implement conversation memory with [memory_window: e.g., last 5 turns]
- Add guardrails for hallucination detection and PII protection
- Support streaming responses for better UX
- Include evaluation harness with [eval_metrics: e.g., faithfulness, answer_relevancy, context_precision]
- Make configuration fully parameterized via YAML/JSON config file
</constraints>

<format>Provide the complete template as a structured Python project with:
1. Project structure diagram
2. Configuration schema (config.yaml)
3. Core modules: ingestion, chunking, embedding, retrieval, generation, evaluation
4. Main pipeline orchestration class
5. CLI entry point with commands: ingest, query, evaluate, serve
6. FastAPI/Gradio server implementation
7. Dockerfile and docker-compose.yml
8. Comprehensive README with usage examples
9. Unit and integration tests
10. Example evaluation dataset format</format>

<tone>Technical, precise, engineering-focused, and production-oriented. Use clear code comments and type hints throughout.</tone>

<final_instruction>Generate the complete RAG chat template project structure with all core modules, configuration, and documentation. Begin with the project tree diagram and config.yaml schema.</final_instruction>