Design LlamaIndex RAG Pipeline Architecture

Design a LlamaIndex-based RAG pipeline for processing heterogeneous enterprise documents. Focus on creating custom document loaders for PDF and audio transcripts (e.g., using `pypdf` for PDFs, or simple regex for `.txt` transcripts), defining appropriate chunking strategies for Gemini 2.5 Pro, and outlining the indexing process using MongoDB Atlas Vector Search. Describe how to integrate LlamaIndex's knowledge graph functionality to enrich document understanding. Ensure your architecture can support generating summaries and answering complex queries.