Retrieval-Augmented Generation (RAG) · TRACK 02/05
Chunking & Ingestion
Where most RAG quality is won or lost: how documents become chunks.
// THE TRACK
01 · START HEREChunking in RAGYou'll understand why documents must be split before embedding and how chunk boundaries shape everything downstream in a RAG system.BEGINNERStrategies ComparedYou'll know the real trade-offs between fixed-size, recursive, semantic, and document-level chunking — and how to pick the right one for your corpus and query patterns.INTERMEDIATEChunk Size & OverlapYou'll know sensible starting values for chunk size and overlap, and a simple process for tuning them on your own documents.BEGINNERCode, Tables & MarkdownYou'll know how to chunk source code, tables, and Markdown without destroying the structure that makes them retrievable.INTERMEDIATEParsing PDFsYou'll understand the failure modes of PDF extraction and how to handle tables, multi-column layouts, and scanned pages in a RAG pipeline.INTERMEDIATEContextual RetrievalYou'll learn how prepending document-level context to each chunk fixes the 'orphaned chunk' problem and lifts retrieval accuracy.INTERMEDIATECleaning Data for RAGYou'll get a practical checklist for cleaning messy documents so your RAG system retrieves signal instead of noise.BEGINNERSemantic ChunkingYou'll understand how semantic chunking uses embeddings to split documents at topic boundaries instead of arbitrary character counts.INTERMEDIATEUnstructuredYou will understand what Unstructured does, how it converts messy documents into clean structured elements, and why it is a common first step in a RAG pipeline.INTERMEDIATEDoclingYou will understand what Docling is, how its advanced layout and PDF understanding produces clean Markdown or JSON, and where it fits in a RAG ingestion pipeline.INTERMEDIATELlamaParseYou will understand what LlamaParse is, how VLM/LLM-powered agentic OCR reads complex layouts, and when a hosted parser beats a local library for RAG ingestion.INTERMEDIATEChonkieYou will understand what Chonkie is, the chunking strategies it offers (token, sentence, semantic, LLM), and how a dedicated chunking library improves retrieval.INTERMEDIATE