reddit · SideProject · 10-06 21:50 54 分 ← 返回列表

I built a tool that reads supplier invoices and hands back checked data

I've been building out my document intelligence app DocReadi, an AI-powered document extraction app, with human-in-the-loop features to allow for the accuracy that finance processes require. Think of it as OCR 2.0 AI-edition You send it invoices, receipts, credit notes or supplier statements by email, upload, the API, or a watched SFTP, S3, OneDrive or SharePoint folder. It reads the fields and the line items and checks the arithmetic: the lines have to add up to the subtotal, and subtotal

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：财务/记账场景下，供应商发票、收据、对账单的字段提取要求极高准确率，纯 OCR 或纯 LLM 抽取会编造数字，必须有人工复核环节才能进入后续财务流程。

目标用户：中小企业财务团队、外包记账公司、需要处理大量供应商单据的会计与运营人员

证据：I've been building out my document intelligence app DocReadi, an AI-powered document extraction app, with human-in-the-loop features to allow for the accuracy that finance processes require... The model only reads what's printed, and anything missing is left blank and flagged.

变现：按处理页数/单据量的 SaaS 订阅（免费试用 + 阶梯定价），或按 Xero/QuickBooks 导入文件导出次数收费；可叠加 API 调用与企业版 SFTP/S3 接入溢价。

MVP：两周版 MVP：一个收件邮箱 + 上传入口，用 LLM 抽取发票头与行项目，强制输出"页面上没有"的缺失字段并高亮，做算术校验（行合计=小计、小计+税=总额），加一个审核队列页面，通过后导出 CSV/Xero Bills 导入文件。

竞品：Rossum、Docparser、Nanonets、Mindee、Klippa、AWS Textract、Azure Document Intelligence

摘要：作者发布了自己独立开发的 DocReadi，用 AI + human-in-the-loop 从邮件、API、SFTP/S3/OneDrive/SharePoint 读取供应商发票、收据和信用票据，校验行项目与小计、税额的算术一致性，人工审核后再导出 CSV/Excel/JSON/Xero Bills 或 webhook。产品刻意不做审批、付款、银行流水和对账，定位为现有财务流程前置层。帖子中没有展示任何付费用户或收入数据，需求验证主要来自作者个人判断，且该赛道已有 Rossum、Nanonets、Mindee 等成熟竞品。