documentai 开源项目推荐（17 个）

mindee /doctr

docTR（文档文本识别）——一个由深度学习驱动的、用于OCR相关任务的无缝、高性能且易用的库。由 t2k 持续开发和维护。

katanaml /sparrow

使用 ML、LLM 和 Vision LLM 进行结构化数据提取、指令调用与智能体工作流

AlibabaResearch /AdvancedLiterateMachinery

面向 Advanced Literate Machinery 的原创、创新想法与算法集合。本项目由阿里巴巴集团通义实验室语言技术实验室 OCR 团队维护。

robocorp /rpaframework

Collection of open-source libraries and tools for Robotic Process Automation (RPA), designed to be used with both Robot Framework and Python

docprompt /Docprompt

Enterprise-ready Document Analysis with Large Language Models

landing-ai /ade-docvqa-benchmark

99.156% Accuracy from Agentic Document Extraction DPT-2 model on DocVQA val split

0xthierry /layout-document-ai

Document AI preserving the original layout.

amannanda-22 /DOCMINDPRO

Production-grade Multi-PDF RAG Intelligence System with FastAPI, Streamlit, LangChain, Gemini, ChromaDB, BM25, CrossEncoder, and RAGAS.

oddell /Invoice-Processor

AI Paper or PDF Invoices to Excel Consolidation 📄 ➡️ 🧠 ➡️ 📁

SUGUKURU-CO-LTD /mcp-regulated-cookbook

Production patterns for running Model Context Protocol servers in regulated business domains. Apache 2.0.

juliusdejon /expense-tracker

Receipts to texts using NodeJS and Google Document AI

srinandan /building-permit

An Agentic application to review building plans against the California Building Standards Code and Santa Clara County reach codes

dhruvakamra /documentai

Google document AI automation artifacts using python

ondongo /artisan

Analyse de devis via IA

piyushxbhardwaj /DeltaDocAI

AI-powered engineering document comparison with delta analysis, grounded RAG chat, OCR, and visual diff generation.

Kalyx1111 /DocRAG

DocRAG is a fully offline, private document intelligence assistant that lets you permanently store and query PDF, DOCX and TXT files. Powered by Qwen2.5 and bge-m3 through llama.cpp, it searches your entire personal library and provides citation-grounded answers with source and page references.

elenafmoseyko /claims-document-ai-pdf-extract

Document AI pipeline that extracts structured claims data from PDF reports and outputs formatted Excel exhibits, and JSON data.