When I first talked about my Part 1 of this journey, the assistant could only detect passive voice—covering roughly 30% of the most common style issues in technical documentation. It was helpful—but limited.

Now, it’s so much more.

From detecting a single issue to handling five critical style rules.

The Vision: An AI That Understands Documentation Style

General grammar checkers aren’t built for technical content. This AI assistant is.

It’s trained on thousands of real documentation corrections made by writers, editors, and reviewers—covering rules like:

  • Passive to active conversion
  • Redundancy reduction
  • Tone normalization
  • Wordy phrase rewrites
  • Ambiguity elimination

“The goal isn’t to replace reviewers—it’s to handle the repetitive, rule-based corrections automatically, so humans can focus on what matters.”

What’s Inside?

Here’s a snapshot of how far this has come:

Evolution of the AI Writing Assistant

Feature Initial Prototype
(April 2025)		 Current Release
(May 2025)		 Impact
Rules Coverage Passive voice only 5 critical style rules +400% rule coverage
Dataset Size 2,500 examples 15,000+ curated pairs 6x more training data
Output Style Line-by-line corrections Context-aware paragraph rewrites More cohesive document flow
Document Support Plain text only
.txt		 Multiple formats
.md, .adoc, .docx, .pdf		 Works with real-world documentation
Accuracy ~80% for passive voice ~92% across all rules +12% detection accuracy
Processing Speed 3-5 sec/paragraph 0.8 sec/paragraph 5x faster processing
Integration Standalone tool CI/CD pipeline compatible Fits into existing workflows
Domain Adaptation Generic writing rules Technical documentation focused Context-aware suggestions
UI Access Command line only Gradio UI + API access Accessible to non-technical users

Try It Yourself

You can test the assistant in your own workflow with:

Future Development Roadmap

Phase 1: Advanced Applications Beyond Style

Moving forward, I plan to extend the model to handle more complex documentation challenges. The techniques I’m exploring in my AI/ML Documentation Course will be particularly valuable for training the model to understand vector embeddings and knowledge graphs within technical content.

The goal is to create an assistant that can intelligently process API specifications and ML model documentation, identifying not just style issues but also inconsistencies in technical accuracy. By applying the same training methodology developed for documentation style, we envision specialized models for validating API endpoint descriptions against OpenAPI schemas or ensuring ML model behavior is accurately documented.

Phase 2: Extreme Rule Enforcement for Minimalist Documentation

The future iterations of this model will focus on enforcing even more sophisticated writing rules, particularly ones that separate merely good documentation from truly exceptional content:

  1. Future Tense Elimination - Converting all future tense statements (“will happen”) to present tense (“happens”) for more direct, authoritative documentation

  2. Aggressive Minimalism - Identifying and removing every unnecessary word, aiming for the absolute minimum word count without losing meaning

  3. Technical Precision Enhancement - Detecting vague technical descriptions and replacing them with exact, measurable statements

  4. Contextual Word Choice - Understanding domain-specific terminology and suggesting the most appropriate terms based on the technical context

These advanced rules build on the foundation established in the current release but push the boundaries of what’s possible with AI-assisted technical writing. The techniques explored in my API Documentation and AI/ML Documentation courses will be further refined to train these more sophisticated rule sets.

Previous Article

Building an AI Technical Writing Assistant - Transforming Documentation with Machine Learning

Learn how to create an AI writing assistant that catches passive voice, elimi...