Hierarchical Markdown Structure for Vector Embeddings: Messy document layouts destroy RAG retrieval accuracy. In this project, I designed a strict Markdown heading hierarchy (#, ##, ###) specifically optimized for text chunking. By programmatically maintaining parent-child document relationships, I ensured that semantic search algorithms in the vector store can accurately retrieve the exact context an LLM needs, drastically reducing hallucinations.