Build LlamaIndex RAG System

Build a production-ready LlamaIndex RAG system for [PROJECT_NAME] with the following specifications:

Requirements

• Project name: [PROJECT_NAME]
• Document sources: [DIRECTORY_PATH or FILE_PATHS]
• Document types: [PDF/TXT/DOCX/MD]
• LLM provider: [OpenAI/Anthropic/Local]
• Model name: [gpt-4/claude-3/llama-2]
• Embedding model: [text-embedding-ada-002/custom]
• Vector store: [Default/Pinecone/Weaviate/Chroma]
• Metadata fields: [FIELD_1], [FIELD_2], [FIELD_3]
• Query modes: [Simple Q&A/Chat/Streaming]
• Persistence needed: [YES/NO]
• Retrieval strategy: [Similarity/MMR/Rerank]

Deliverables

Generate complete LlamaIndex RAG system with:

Document loading and processing:

• SimpleDirectoryReader or custom loader for [DOCUMENT_TYPES]
• Document metadata extraction for [METADATA_FIELDS]
• Text splitter configuration for optimal chunk sizes
• Document preprocessing and cleaning

Vector index creation:

• VectorStoreIndex setup with [VECTOR_STORE]
• Custom ServiceContext with [LLM_PROVIDER] and [MODEL_NAME]
• Embedding configuration for [EMBEDDING_MODEL]
• Index persistence to [STORAGE_PATH] if needed

Query engine setup:

• Query engine with [RETRIEVAL_STRATEGY]
• Response synthesis mode configuration
• Custom prompts for [PROJECT_NAME] use case
• Similarity threshold and top-k configuration
• Metadata filtering for [METADATA_FIELDS]

Chat engine if needed:

• Conversational interface with context retention
• Chat mode selection (condense_question/context/refine)
• Chat history management
• Streaming responses if required

Advanced features:

• Custom prompt templates for domain-specific responses
• Response evaluation and feedback loop
• Query transformation for better retrieval
• Sub-question query engine for complex questions

Storage and loading:

• Index persistence implementation
• Loading from storage for fast startup
• Incremental index updates

Output complete, production-ready RAG application with all configuration and usage examples.