IntelliSense AI Writing Assistant

Built with Kiro during the "Code with Kiro" hackathon

Inspiration

As a developer who writes constantly - documentation, emails, code comments, social media posts - I was frustrated by the binary choice between overwhelming writing tools like Grammarly and basic spell checkers that miss nuanced errors. I wanted something that could adapt to different contexts: sometimes I just need quick spelling fixes during casual messaging, other times I want comprehensive AI assistance for professional writing.

The inspiration struck when I realized that writing assistance should be contextual and adaptive. A developer debugging code needs different help than someone crafting a business proposal. This led to the core concept: preset modes that match how people actually write in different situations.

But the real inspiration came from discovering Kiro's potential. Instead of building another generic writing tool, I wanted to explore how AI-assisted development could help create something truly sophisticated - a Chrome extension that works seamlessly across every website with intelligent, real-time assistance.

What it does

IntelliSense AI Writing Assistant is a Chrome extension that provides intelligent, context-aware writing assistance across the entire web. It features three distinct preset modes:

🔤 Basic Mode - Seamless Spelling Correction

Instantly fixes spelling mistakes while preserving formatting and style
Non-intrusive corrections that happen as you type
Perfect for casual messaging and quick communications

🎯 Minimalistic Mode - Grammar Perfection

Intelligent grammar correction that maintains your voice and tone
Fixes subject-verb agreement, tense consistency, and sentence structure
Ideal for professional emails and formal documents

🚀 Full Mode - Complete AI Writing Experience

Smart Autocomplete: Context-aware word and phrase completions with 3 options
Thinking Mode: After completing sentences, offers intelligent continuations
Sequential Processing: Grammar correction → Autocomplete → Sentence continuations with natural 3-second delays
Real-time Assistance: All features working together seamlessly

Core Features

Universal Compatibility: Works on Gmail, Google Docs, LinkedIn, Twitter, Facebook, and thousands of other websites
70+ Language Support: Automatic language detection with AI-powered corrections
Smart Caching: Intelligent response caching with 30-minute expiry for instant performance
Keyboard Navigation: Full accessibility with Tab, Escape, and arrow key support
Security First: Automatically skips password fields and sensitive inputs

How we built it

Building IntelliSense was a journey of architectural complexity that Kiro helped navigate from start to finish.

The Kiro-Driven Development Process

Instead of diving into code, I started with Kiro's spec system. This was transformative - breaking down the massive feature set into manageable, structured tasks:

# Improved Language Detection Spec

- [x] 1. Add language detection AI method
- [x] 2. Implement language detection caching system
- [x] 3. Add debug logging for language detection
- [x] 4. Simplify grammar correction prompt
- [x] 5. Integrate language detection into grammar correction flow

Architecture Collaboration with Kiro

Multi-Layer Architecture:

Background Layer: Service worker managing Groq API calls, caching, and rate limiting
Content Layer: Text monitoring and UI injection across all websites
UI Layer: Context-aware interfaces for three preset modes
API Layer: Sequential AI processing with intelligent language detection

Key Technical Decisions Made with Kiro:

TextMonitor Class Design: Kiro helped architect a robust system for monitoring user input across diverse websites:

class TextMonitor {
  constructor() {
    this.debounceDelay = 500; // Smart input debouncing
    this.thinkingModeDelay = 1500; // Natural pause detection
    this.activeElements = new Set(); // Efficient element tracking
    this.lastTextStates = new Map(); // State management
  }
}

Sequential Processing Flow: Kiro guided the implementation of natural AI assistance timing:

async startSequentialProcessing(element, text, elementId) {
  // Step 1: Grammar correction
  await this.requestAutomaticGrammarCorrection(element, text);
  await this.delay(3000); // Natural pause

  // Step 2: Autocomplete for partial words
  const lastWord = this.getLastPartialWord(updatedText);
  if (lastWord && lastWord.length > 2) {
    await this.requestAutocomplete(element, updatedText, lastWord);
    await this.delay(3000);
  }

  // Step 3: Sentence completion/continuation
  await this.handleSentenceCompletion(element, finalText);
}

Technology Stack

Frontend: React 18 with Webpack 5 for modern development experience
Extension: Chrome Manifest V3 for latest security and performance standards
AI Engine: Groq API with Llama3-8B-8192 model for fast, accurate responses
Languages: JavaScript/TypeScript hybrid approach for flexibility
Build System: Custom Webpack configuration with hot reload for development

Kiro's Code Generation Magic

The breakthrough moment came when I needed cross-site compatibility. I described the challenge to Kiro: "Monitor text input across thousands of websites with different element structures, handle dynamic content, avoid security issues."

Kiro generated a comprehensive solution:

getTextInputSelectors() {
  return [
    'input[type="text"]', 'input[type="email"]', 'input[type="search"]',
    'textarea', '[contenteditable="true"]', '[role="textbox"]',
    '.ql-editor', // Quill editor
    '.DraftEditor-editorContainer', // Draft.js
    '.notranslate', // Google Docs
    // Comprehensive coverage generated by Kiro
  ];
}

Challenges we ran into

Technical Challenges

1. Cross-Site Compatibility Nightmare Every website implements text inputs differently. Gmail uses contenteditable divs, Google Docs has complex nested structures, social media sites use custom components. Kiro helped create a universal monitoring system that works across all of them.

2. Performance Optimization Real-time AI assistance could easily overwhelm both the API and the user experience. We implemented:

Smart debouncing with 500ms input delays
Intelligent caching with LRU eviction
Rate limiting with 3-second delays between API calls
Efficient DOM monitoring with MutationObserver

3. Sequential Processing Complexity The biggest challenge was making AI assistance feel natural, not robotic. Users don't want instant corrections - they want thoughtful, timed assistance. Kiro helped design a system where:

Grammar correction happens after brief pauses
Autocomplete appears for partial words
Sentence continuations offer after completing thoughts
Each step has natural delays to avoid overwhelming users

4. Language Detection Accuracy Initially used regex-based language detection, which failed for mixed content and short texts. Kiro guided the implementation of AI-powered detection:

async detectLanguageWithAI(text) {
  const completion = await this.groq.chat.completions.create({
    model: 'llama3-8b-8192',
    messages: [{
      role: 'system',
      content: 'Detect the language of the following text. Respond with only the language name in English, nothing else.'
    }, {
      role: 'user',
      content: text
    }],
    max_tokens: 10,
    temperature: 0.1
  });

  return completion.choices[0]?.message?.content?.trim() || 'English';
}

User Experience Challenges

5. Non-Intrusive UI Design Writing assistance should enhance, not interrupt. We designed interfaces that:

Appear only when needed
Match the visual style of each website
Disappear gracefully when dismissed
Support full keyboard navigation

6. Preset Mode Complexity Balancing three different user experiences while maintaining code simplicity required careful architecture. Kiro helped create a clean settings system:

applyPresetMode(mode) {
  switch (mode) {
    case 'basic':
      this.settings = { showSpellingCorrection: true, /* others: false */ };
      break;
    case 'minimalistic':
      this.settings = { showGrammarCorrections: true, /* others: false */ };
      break;
    case 'full':
      this.settings = { /* all features: true */ };
      break;
  }
}

Development Process Challenges

7. Architectural Decision Paralysis With so many moving parts, it was easy to get overwhelmed by architectural decisions. Kiro's spec system provided structure, breaking complex features into manageable tasks with clear dependencies.

8. Chrome Extension Complexity Manifest V3 restrictions, content script limitations, and background service worker constraints created a complex development environment. Kiro helped navigate these constraints while maintaining feature richness.

Accomplishments that we're proud of

Technical Achievements

1. Universal Web Compatibility IntelliSense works seamlessly across thousands of websites - from Gmail and Google Docs to social media platforms and custom web applications. This required solving complex DOM manipulation challenges that Kiro helped architect elegantly.

2. Intelligent Sequential Processing The crown jewel is our sequential AI processing system. Instead of bombarding users with simultaneous corrections and suggestions, we created a natural flow:

Grammar correction → Wait 3 seconds → Autocomplete → Wait 3 seconds → Sentence continuations
Feels like having a thoughtful writing partner, not an aggressive autocorrect

3. Advanced Language Detection Moving beyond regex-based detection to AI-powered language identification that works with mixed content, short texts, and 70+ languages. The system maintains 95%+ accuracy while being fast enough for real-time use.

4. Smart Caching Architecture Implemented a sophisticated caching system that:

Reduces API calls by 60% through intelligent response caching
Uses LRU eviction to manage memory efficiently
Provides instant responses for repeated corrections
Maintains 30-minute cache expiry for freshness

User Experience Achievements

5. Three Distinct Preset Modes Created genuinely different user experiences:

Basic: Seamless spelling correction that's invisible until needed
Minimalistic: Professional grammar assistance without distractions
Full: Complete AI writing partnership with all features

6. Thinking Mode Innovation The "thinking mode" feature detects when users complete sentences and pause, then offers intelligent continuations. This feels natural and helpful rather than intrusive.

7. Accessibility Excellence Full keyboard navigation support with Tab, Escape, and arrow keys. The extension works seamlessly with screen readers and other assistive technologies.

Development Process Achievements

8. Kiro-Driven Architecture Successfully used Kiro's spec system to manage complex feature development. The structured approach led to cleaner code, fewer bugs, and faster development cycles.

9. Production-Ready Quality Built a Chrome extension that's not just a hackathon demo, but a production-ready tool with:

Comprehensive error handling
Security best practices
Performance optimization
Professional UI/UX design

What we learned

About AI-Assisted Development

1. Specs Are Game-Changers Kiro's spec system transformed how I approach complex projects. Instead of diving into code, starting with structured task breakdowns led to better architecture and fewer rewrites.

2. Collaborative Problem-Solving The most powerful aspect wasn't code generation, but having conversations about solutions. Kiro helped me think through problems systematically, considering edge cases and architectural implications I might have missed.

3. Iterative Development Excellence Working with Kiro encouraged building incrementally. Each feature built on previous work, creating a solid foundation rather than a house of cards.

Technical Insights

4. Chrome Extension Architecture Learned the intricacies of Manifest V3, content script limitations, and background service worker constraints. Kiro helped navigate these complexities while maintaining feature richness.

5. Real-Time AI Integration Discovered the importance of intelligent timing in AI assistance. Users don't want instant responses - they want thoughtful, well-timed help that enhances their natural writing flow.

6. Cross-Site Compatibility Strategies Developed robust techniques for working across diverse websites with different DOM structures, CSS frameworks, and JavaScript implementations.

User Experience Lessons

7. Context-Aware Design Different writing contexts require different assistance levels. The preset mode approach proved that one-size-fits-all solutions don't work for writing tools.

8. Non-Intrusive AI The best AI assistance is invisible until needed. Users want enhancement, not replacement of their writing process.

Performance Optimization

9. Caching Strategy Importance Smart caching reduced API costs by 60% while improving user experience. The key was balancing cache freshness with performance gains.

10. Debouncing and Rate Limiting Learned that aggressive real-time processing can overwhelm both APIs and users. Strategic delays and rate limiting created a more natural experience.

What's next for IntelliSense AI Writing Assistant

Immediate Roadmap (Next 3 Months)

1. Chrome Web Store Launch

Complete final testing and optimization
Prepare store listing with screenshots and demo videos
Implement user feedback collection system
Launch public beta with limited user base

2. Advanced Language Features

Expand to 100+ languages with regional dialect support
Add language-specific writing style corrections
Implement cultural context awareness for international users
Add translation assistance between languages

3. Performance Enhancements

Implement local AI models for basic corrections (offline capability)
Add predictive caching based on user writing patterns
Optimize memory usage for long writing sessions
Reduce extension size through code splitting

Medium-Term Goals (3-6 Months)

4. Advanced AI Features

Voice Input Integration: Speech-to-text with real-time correction
Writing Style Analysis: Detect and maintain consistent tone across documents
Content Suggestions: AI-powered topic and structure recommendations
Plagiarism Detection: Basic originality checking for academic and professional writing

5. Team and Collaboration Features

Shared Style Guides: Team-wide writing standards and corrections
Collaborative Editing: Real-time suggestions for shared documents
Writing Analytics: Personal improvement tracking and insights
Custom Dictionaries: Industry-specific terminology and corrections

6. Platform Expansion

Firefox Extension: Adapt architecture for Mozilla's extension system
Safari Extension: iOS and macOS support for Apple ecosystem
Mobile Support: Progressive Web App for mobile browsers
Desktop Integration: Native apps for Windows, macOS, and Linux

Long-Term Vision (6-12 Months)

7. Advanced AI Capabilities

Custom Model Training: User-specific writing style learning
Context-Aware Suggestions: Understanding document purpose and audience
Multi-Document Consistency: Maintaining style across related documents
Advanced Grammar Rules: Industry-specific and academic writing standards

8. Integration Ecosystem

Google Workspace Integration: Deep integration with Docs, Gmail, and Sheets
Microsoft Office Support: Word, Outlook, and Teams compatibility
Slack and Discord Bots: Real-time assistance in team communications
API for Developers: Allow third-party integrations and custom implementations

9. Enterprise Features

Admin Dashboard: Team usage analytics and configuration management
Compliance Tools: Industry-specific writing standards enforcement
Security Enhancements: On-premise deployment options for sensitive environments
Custom Branding: White-label solutions for enterprise clients

Research and Innovation

10. Next-Generation Features

Emotional Intelligence: Tone and sentiment analysis with suggestions
Visual Writing: Integration with design tools for content creation
Code Documentation: Specialized assistance for technical writing
Academic Research: Citation management and research paper assistance

Community and Open Source

11. Developer Community

Open Source Components: Release core modules for community contribution
Plugin Architecture: Allow developers to create custom correction rules
API Documentation: Comprehensive guides for third-party integrations
Developer Grants: Fund community-driven feature development

12. User Community

Beta Testing Program: Early access to new features for power users
Feature Request System: Community-driven development priorities
Writing Workshops: Educational content and best practices sharing
Success Stories: Showcase how users improve their writing with IntelliSense

Technical Architecture Evolution

The current architecture provides a solid foundation for these expansions:

Current: Browser Extension → Groq API → Real-time Assistance
Future: Multi-Platform App → Hybrid AI (Local + Cloud) → Comprehensive Writing Suite

Scalability Considerations:

Microservices architecture for different AI capabilities
Edge computing for reduced latency
Progressive enhancement for offline functionality
Modular design for platform-specific optimizations

AI Model Evolution:

Transition from single Groq model to specialized model ensemble
Custom fine-tuning for specific writing domains
Integration with multiple AI providers for redundancy
Local model deployment for privacy-sensitive users

IntelliSense AI Writing Assistant represents just the beginning of what's possible when AI-assisted development meets genuine user needs. Built with Kiro's guidance, it demonstrates how thoughtful AI partnership can create tools that truly enhance human capabilities rather than replace them.

The journey from hackathon project to production-ready extension has been transformative, both for the software and for my development approach. With Kiro as a development partner, the future roadmap feels not just achievable, but exciting.

Project Repository: https://github.com/lxjeman/intellisense

Built with ❤️ and Kiro during the "Code with Kiro" hackathon

Built With

google-extention
javascript
kiro
kiroide

Updates

Valeriu Jeman posted an update — Aug 22, 2025 06:32 AM EDT

App new Site - https://intellisense.vercel.app/

Log in or sign up for Devpost to join the conversation.

Valeriu Jeman started this project — Jul 27, 2025 04:03 PM EDT

Leave feedback in the comments!