Multimodal RAG Pipeline Design

Design a multimodal RAG pipeline supporting text, images, and documents.

## Content Types
{{content_types}}

## Query Types
{{query_types}}

## Integration Requirements
{{integration_requirements}}

Design the pipeline:

**Ingestion Layer**
- Document parsing (PDF, DOCX)
- Image extraction and captioning
- Table/chart understanding
- Unified metadata schema

**Embedding Layer**
- Text embedding model
- Vision embedding model
- Multimodal alignment

**Retrieval Layer**
- Cross-modal search
- Modality-specific filtering
- Result fusion

**Generation Layer**
- Multimodal context assembly
- Vision-language model integration
- Citation with visual references

Provide architecture diagrams and key implementation code.