Before Full Agentic RAG: Know How You Decide, and the Parsing Methods You Pick From
| Source: Towards Data Science
Tags: RAG, document parsing, enterprise AI, Docling, PaddleOCR, agentic AI, LLM
A detailed technical guide builds a controlled enterprise RAG document dispatcher that reads each PDF's nature and selects the right parser (fitz, Docling, PaddleOCR, EasyOCR, MinerU, or Surya) — challenging the industry's loose use of 'agentic' for what are actually deterministic rule-based pipelines.
Details
This article closes the document parsing layer of a four-part Enterprise Document Intelligence series. The core pattern is a dispatcher that reads a PDF's nature, plans an ordered set of methods, executes them, and folds outputs into one enriched corpus. The parsing toolkit covers six scenarios: fitz for native text layers, Azure Document Intelligence and Docling for tables, vision LLMs for charts and diagrams, EasyOCR for scanned pages with no text layer, image captioning for skipped content, and two strategies for recovering table of contents. The key intellectual contribution is the argument against 'agentic' labeling: most vendor pipelines marketed as agentic are rule-based dispatchers where LLMs operate only inside individual method leaves. For enterprise RAG specifically, this distinction matters — every decision must be explicit and logged because outputs feed real business decisions. The dispatcher is logged and auditable, with each step readable before execution. A runnable notebook on the 1706.03762v7.pdf (Attention Is All You Need paper) is available in the doc-intel/notebooks-vol1 repository.