RAG System Multimodal Document Processing Solution Designer

You are a senior RAG architect specializing in multimodal document processing. I need you to design a comprehensive RAG pipeline for my use case. ## My Requirements: - Document types: [PDF/Word/HTML/images - specify yours] - Content modalities: [text, tables, charts, equations, images - specify yours] - Query types: [factual lookup / analytical / comparison - specify yours] - Scale: [number of documents, average size] ## Please provide: 1. **Document Ingestion Pipeline**: How to parse and chunk multimodal documents while preserving cross-modal relationships 2. **Embedding Strategy**: Which embedding models to use for each modality, and how to align them in a shared vector space 3. **Retrieval Architecture**: Hybrid retrieval design combining dense vectors + sparse keywords + knowledge graph edges 4. **Context Assembly**: How to reconstruct rich context from retrieved chunks before feeding to the LLM 5. **Evaluation Framework**: Metrics and test cases for measuring retrieval quality and answer faithfulness across modalities For each component, provide recommended open-source tools, key configuration parameters, common pitfalls, and a minimal working code snippet in Python. Format your response as a structured technical design document with diagrams described in Mermaid syntax.