← Back to the summary

DocQuality: An Open-Source AI Tool That Audits Word Documents in Seconds

As everyone knows, I've been running a startup these past few years, which inevitably involves handling various contracts, official documents, bidding documents, and other Word files. The review process has long relied on manual checking. Just reviewing a single corporate service contract could take me 1-2 hours, and occasionally, some loopholes would still slip through.

For an AI entrepreneur, efficiency is king, and the situation above was something I couldn't tolerate.

So I deeply reviewed the three core pain points of past manual reviews:

First, extremely low efficiency. Word-by-word proofreading of long documents takes hours, with extremely high labor costs.

Second, poor accuracy. Manual reviews easily miss formatting errors, compliance loopholes, and logical contradictions, relying on experience without a unified standard.

Third, no quantitative standards. Document quality is judged entirely subjectively, making it impossible to produce standardized review reports, which hinders team retrospectives and compliance archiving.

Based on these pain points, I developed an intelligent document review tool — DocQuality.

It is a lightweight, privatizable, out-of-the-box open-source AI document quality review tool that completely overturns the inefficient model of traditional manual document proofreading and compliance review. Using conversational AI + a rules engine, it achieves full-dimensional document quality inspection in tens of seconds, perfectly adapting to various document review scenarios in the workplace, enterprise office, legal compliance, and more.

As usual, here is the open-source address first:

https://github.com/jitOffice/DocQuality

Below, I will introduce this open-source project comprehensively, from its functions to its technical implementation.

Just upload a Docx file and complete a content review with one sentence

The core value of DocQuality is reconstructing the document quality inspection process with AI technology. It abandons the rigid "one-click scan" mode of traditional tools and innovatively adds conversational interaction capabilities. We simply upload a docx document and input review requirements in natural language, and the platform combines a large language model with a self-developed rules engine to complete a deep quality inspection from multiple dimensions, outputting quantitative scores, problem lists, and precise modification suggestions, completely shifting document review from "manual experience-driven" to "AI standardization-driven."

Here is the result after the review is completed:

In the dialog box, we can see the quality and compliance of the Word document at a glance, and also preview the docx document online with one click and view the complete analysis report:

For the online preview of Word documents, I used the jitword collaborative document engine, which enables high-precision Word file previewing and editing.

Through conversational review, we can complete work that originally required several hours of manual effort in under 5 minutes, greatly improving office efficiency.

A three-dimensional, full-scenario quality inspection engine covering most document review needs

I designed a three-dimensional quality review system that can comprehensively inspect document risks, covering 99% of office document quality inspection needs:

Format Specification Review: Automatically detects basic issues such as chaotic heading levels, numbering errors, punctuation mistakes, and inconsistent formatting, suitable for formal document format verification like official documents and reports;
Compliance Risk Review: Accurately identifies sensitive private information, legally risky expressions, missing essential clauses, unfilled placeholders, and other issues, suitable for compliance verification of contracts, bids, and marketing copy;
Content Consistency Review: Intelligently checks for inconsistent terminology, logical contradictions, repetitive or redundant content, and incoherent sentences, significantly improving document professionalism.

It also comes with a complete quantitative scoring system, using a hundred-point total score and a four-level rating (Excellent/Good/Fair/Poor), breaking down sub-scores across three dimensions, making document quality clear at a glance with more intuitive and professional results.

Let's look at the review report generated by DocQuality:

The three-dimensional analysis content of the review report is as follows:

Intelligent Scoring

Review Process Details

Review Detail Information

In this way, we can optimize contract content more efficiently.

A flexible and extensible rules engine, adapting to various document review needs

DocQuality balances ease of use for novice users with professionalism for expert users. It has built-in rule templates for 6 major industry scenarios, covering high-frequency scenarios such as advertising compliance, privacy protection, and contract review:

We can add them to the rule panel with one click, and they will automatically take effect the next time a document review is performed.

On the other hand, I also provided a custom rule feature, allowing users to directly use the keyword configuration mode without needing to master complex regular expressions:

For more professional users, they can directly switch to the advanced regex mode, supporting custom complex verification rules:

It also supports rule import, export, and secondary editing, offering strong extensibility.

After configuring the rules, we can select specific rules for conversational review on the intelligent review page:

Multi-model compatibility + full-process report collaboration

DocQuality supports mainstream large models such as DeepSeek, Tongyi Qianwen, Kimi, and OpenAI, while also being compatible with all custom models following the OpenAI protocol. We can freely configure the API endpoint and model parameters. I also built in API key availability detection, supporting one-click model switching.

After the review is complete, it supports one-click export of reports in HTML, PDF, and Word formats. Historical records are traceable, comparable, and previewable, and can also be exported as CSV files with one click.

Paired with JitWord's online document preview capability, we can check the original text without downloading, adapting to various scenarios such as team collaboration and retrospectives.

Technical Architecture and Tech Stack Implementation

DocQuality's technical architecture is divided into five core layers: the frontend interaction layer, backend service layer, AI capability layer, document parsing layer, and data storage layer. The architecture design is as follows:

Here, I'll also share the tech stack for your reference:

Frontend: React 18, TypeScript, Vite, TailwindCSS, Zustand, React Router v6
Backend: NestJS, TypeORM, SQLite, Axios
Engineering: pnpm workspace, Monorepo, concurrently
AI Capability: OpenAI-compatible protocol, multi-model adaptation
Document Service: JitWord structured parsing service

If you are interested in the project, you can refer to the source code on GitHub for research:

https://github.com/jitOffice/DocQuality

Application Scenario Analysis

DocQuality's capabilities are highly versatile, and its rule configuration supports customization, adapting to the vast majority of government, enterprise, and personal office document quality inspection scenarios. Based on my own experience, I've listed a few application scenarios for your reference:

Contract Legal Review: Automatically detects legal risks such as missing liability clauses, unfilled placeholders, and non-compliant wording in contracts, reducing the pressure on legal review;
Official Document Format Proofreading: Standardizes the verification of heading levels, numbering, punctuation, and formatting in official documents, ensuring document standardization and uniformity;
Bidding Document Quality Inspection: Checks bid documents for sensitive information leaks, completeness of essential clauses, and content consistency, mitigating bidding risks;
Marketing Copy Compliance: Accurately identifies prohibited superlatives and non-compliant expressions under advertising law, avoiding compliance risks in marketing content;
Work Report Quality Inspection: Corrects issues like chaotic terminology, logical contradictions, and repetitive redundancy in documents, enhancing report professionalism.

A final small summary

Creating this open-source project was entirely based on the actual pain points of our current startup journey, hoping to help enterprises improve human efficiency and fully enhance employee document review efficiency.

Unlike generic AI document tools on the market, DocQuality focuses on the vertical scenario of quality review, without redundant features. Its advantages lie in being lightweight, easy to deploy, and suitable for private enterprise document scenarios. Moreover, enterprises can perform secondary development based on it to meet complex business needs in various scenarios.

We will continue to iterate in the future. If you have any good suggestions or ideas, feel free to leave feedback in the comments section!

github: https://github.com/jitOffice/DocQuality

Open source is not easy, everyone please give it a star, and contributions are welcome~

That's all for now. I will continue to share notes on AI entrepreneurship and open source in the future. Welcome to leave comments and exchange ideas~