pdf-parsing

The author is working on a Credit Underwriting AI Agent and is seeking open-source on-premise approaches to extract structured transaction tables from diverse Indian bank statement PDFs, facing challenges with layout variations and needing reliable extraction methods.

A benchmark comparing 8 PDF parsers across 14 capabilities, finding Chandra the most accurate while noting trade-offs like speed; LightOnOCR-1B impresses for its size but hallucinates on illegible text.

LiteParse now supports extracting structured data from PDFs—form fields, checkbox states, annotations, images, vector graphics, and word-level bounding boxes—without a vision model, plus complexity signals to route harder pages to tools like LlamaParse.

A developer benchmarks three PDF-parsing models (MinerU, Granite-Docling, PaddleOCR-VL) across six document types and twelve capabilities, finding MinerU drops footers unless its markdown is rebuilt while Granite-Docling outputs cleaner native markdown tables.

The author explains why a customer's PDF bank statement showed all amounts as negative: the PDF encodes invisible trailing minus signs in a gray level matching the background, which the extraction code reads as visible text. They discuss solutions like OCR or stripping non-black text.

Vik Paruchuri announces that his PDF parsing tool Marker has grown into Datalab projects with 72k GitHub stars and 6M monthly downloads, and they are hiring a founding open-source engineer.

LiteParse is a fast, lightweight, open-source PDF parsing tool written in Rust, supporting image-to-PDF conversion natively and providing spatial text parsing with bounding boxes, available via multiple packages (Rust, Node.js, Python, WASM).

LlamaIndex and LanceDB collaborated on a pipeline using LiteParse for PDF parsing and LanceDB for multimodal storage, enabling better retrieval from complex enterprise PDFs for agentic workflows.

Introduces five open-source tools (MarkItDown, MinerU, Docling, marker, surya) that convert PDFs, Office documents, etc., into Markdown or JSON for direct use with RAG or AI agents.

Built a fast PDF parsing service using LiteParse, an open-source tool by LlamaIndex, with help from Cursor AI.

LiteParse is a fast, open-source document parser that outperforms some frontier VLMs on markdown parsing without using AI models. It is available in multiple languages and platforms, and is developed by LlamaIndex.

LlamaIndex improved their LiteParse PDF parsing skill for Claude agents, making it 37% cheaper and more accurate by optimizing agent behavior through evaluation traces.

LlamaIndex's blog post describes building a custom LiteParse skill for Claude agents that reduced cost per question by 37% and improved answer quality by analyzing agent traces to fix inefficiencies in PDF parsing.

LiteParse is a fast, open-source document parser written in Rust that provides high-quality spatial text extraction with bounding boxes, supporting multiple languages and platforms for AI document workloads.

Parse-Flow is an open-source visual workflow designer built by LlamaIndex that chains four document processing primitives—Parse, Classify, Split, and Extract—into a drag-and-drop canvas powered by LlamaAgents workflows, enabling reliable structured data extraction from unstructured enterprise documents like PDFs, contracts, and invoices.

Jerry Liu of LlamaIndex announces a $1M+ billboard campaign in SF promoting their PDF parsing service for AI agents, and lists their booth appearances at upcoming tech conferences.

LlamaIndex demonstrates how to automate a loan underwriting pipeline using LlamaParse to extract structured data from financial PDFs, with cross-document analysis and human-in-the-loop review.

OpenDataLoader-PDF is an open-source PDF parsing tool that achieves a high accuracy rate of 0.907 in tests with real academic papers. It efficiently converts complex PDF documents (including tables, formulas, and scanned images) into Markdown and JSON, making it ideal for local knowledge bases and RAG applications.

OpenDataLoader is an open-source tool that converts PDFs into structured Markdown and JSON, supporting local processing speeds of up to 100 pages/second without requiring a GPU or incurring API costs, designed specifically for RAG pipelines and PDF accessibility automation.

Jerry Liu discusses challenges with using Vision Language Models for PDF parsing, particularly around ensuring text correctness and maintaining proper reading order while avoiding hallucinations.