Implement Custom LlamaIndex Loaders & Indexing

Implement the custom document loaders and indexing logic for your LlamaIndex application. You'll need to parse PDF content (using libraries like `pypdf`) and simple markdown transcripts. Initialize LlamaIndex and configure `ServiceContext` for Gemini 2.5 Pro (e.g., `Settings.llm = Gemini(model="gemini-pro", api_key="YOUR_KEY")`). Set up your `VectorStoreIndex` using `MongoDBAtlasVectorSearch` as the vector store. Provide Python code snippets for initializing `LLM`, `EmbeddingModel`, `VectorStore`, and creating/persisting the index. Use Fireworks AI for embedding generation.