agents-llamaindexlisted

# Data-Centric LLM Applications (Indexing and Retrieval) When the hard part is data, not dialogue, you need an indexing mindset: ingest documents from many sources, structure them into queryable indices, and retrieve precisely what each query needs. This skill covers that data layer. ## Overview The pipeline is ingest → index → retrieve → synthesize. Ingestion handles the messy reality of documents: PDFs, web pages, databases, APIs, each parsed into clean text with metadata. Indexing organizes that text — vector indices for semantic search, keyword indices for exact match, structured indices for tables, summaries for long documents. Query engines combine retrieval strategies and synthesize answers, citing sources. Done well, the model answers from your data instead of its memory. ## When to use - Question-answering over a document collection: manuals, papers, wikis, tickets. - Building RAG where retrieval quality is the bottleneck. - Ingesting heterogeneous sources that need parsing, cleaning, and metadata. - Agentic workflows where the agent needs structured access to documents, not just chat. ## Core concepts - **Document ingestion**: connectors and parsers that turn raw sources into uniform document objects with metadata (source, date, type). Garbage in, garbage out — invest here. - **Node parsing**: splitting documents into chunks (nodes) with overlap, preserving metadata and relationships (parent/child, prev/next). Chunking strategy determines retrieval q