LinguaShift - AI-Powered Corporate Messaging App

LinguaShift is an AI-powered application built on Cloudflare that helps users detect and eliminate jargon from their communications, making messages clearer and more accessible to diverse audiences.

Assignment Requirements Fulfilled

1. LLM (Large Language Model)

• Implementation: Uses @cf/meta/llama-3.1-8b-instruct-fast via Cloudflare Workers AI
• Usage:
  • Jargon detection with contextual analysis
  • Text rewriting with audience-specific customization
• Location: src/index.ts - handleDetect() and handleRewrite() functions

2. Workflow

• Implementation: Cloudflare Workers for request handling and Durable Objects for state coordination
• Architecture:
  • Workers: Handle HTTP requests and coordinate with AI services (src/index.ts)
  • Durable Objects: Manage per-session state and glossary persistence (src/LinguaState.ts)
• Location: Worker entry point in src/index.ts, Durable Object class in src/LinguaState.ts

3. User Input via Chat or Voice

• Implementation: RESTful API endpoints designed for integration with Cloudflare Pages or Realtime
• Endpoints:
  • POST /api/detect - Analyze text for jargon
  • POST /api/rewrite - Rewrite text for specific audiences
  • GET / - Health check
• Frontend Ready: The Worker exposes CORS-enabled APIs that can be consumed by:
  • Cloudflare Pages (static frontend)
  • Cloudflare Realtime (real-time chat interface)
  • Any HTTP client

4. Memory or State

• Implementation: Durable Objects (LinguaState) provide persistent, per-session state management
• Stored State:
  • Glossary: Custom jargon terms per session for personalized detection
  • Preferences: User preferences (audience defaults, tone settings, etc.)
• Location: src/LinguaState.ts - State management using Durable Object storage
• Persistence: State persists across requests using Cloudflare's Durable Objects storage

Architecture

┌─────────────────┐
│   Frontend      │  (Cloudflare Pages/Realtime)
│   (Chat UI)     │
└────────┬────────┘
         │ HTTP/REST
         ▼
┌─────────────────┐
│  Cloudflare     │  ──→  POST /api/detect
│  Worker         │  ──→  POST /api/rewrite
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
    ▼         ▼
┌─────────┐ ┌─────────────┐
│ Workers │ │  Durable    │
│   AI    │ │  Objects    │
│         │ │             │
└─────────┘ └─────────────┘

Features

• Jargon Detection: AI-powered analysis to identify technical terms and jargon in messages
• Smart Rewriting: Automatically rewrite messages for different audiences (technical, non-technical, executives, etc.)
• Customizable Tone: Adapt rewriting tone (neutral, friendly, professional, etc.)
• Persistent Glossary: Store custom terms per session for personalized detection
• Session Management: Track user preferences and glossary per session using Durable Objects

Getting Started

Prerequisites

• Node.js 18+ and npm
• Cloudflare account with Workers AI enabled
• Wrangler CLI installed globally or via npm scripts

Installation

1. Clone the repository:

git clone <repository-url>
cd worker

2. Install dependencies:

npm install

3. Configure Wrangler (if not already configured):

npx wrangler login

Development

Start the development server:

npm run dev
# or
npm start

The Worker will be available at http://localhost:8787 (or the port shown in terminal).

Deployment

Deploy to Cloudflare Workers:

npm run deploy

API Reference

POST /api/detect

Detects jargon in a message using AI analysis.

Request Body:

{
  "message": "We need to refactor the codebase and optimize the API endpoints for better throughput.",
  "sessionId": "user-123"
}

Response:

{
  "terms": [
    {
      "word": "refactor",
      "reason": "Technical jargon",
      "confidence": 0.9
    },
    {
      "word": "throughput",
      "reason": "Technical term",
      "confidence": 0.85
    }
  ],
  "overall_score": "Heavy"
}

POST /api/rewrite

Rewrites a message for a specific audience and tone.

Request Body:

{
  "message": "We need to refactor the codebase and optimize the API endpoints.",
  "sessionId": "user-123",
  "audience": "non-technical stakeholders",
  "tone": "professional"
}

Response:

{
  "rewritten": "We need to reorganize our code system and improve the connection points to make it work better."
}

Glossary Management (via Durable Object)

The Durable Object exposes endpoints for managing glossary terms:

• GET https://dummy/glossary - Retrieve glossary for a session
• POST https://dummy/glossary - Update glossary for a session

Example Glossary Update:

const state = await getState(env, sessionId);
await state.fetch("https://dummy/glossary", {
  method: "POST",
  body: JSON.stringify({
    glossary: [
      { term: "API", definition: "Application Programming Interface" },
      { term: "Refactor", definition: "Restructure code" }
    ]
  })
});

Configuration

The project is configured via wrangler.toml:

name = "cf_ai_linguashift_worker"
main = "src/index.ts"
compatibility_date = "2024-12-01"

[ai]
binding = "AI"

[durable_objects]
bindings = [
  { name = "LINGUA_DO", class_name = "LinguaState" }
]

[[migrations]]
tag = "v1"
new_sqlite_classes = ["LinguaState"]

Testing

Run tests with Vitest:

npm test

Technologies Used

• Cloudflare Workers: Serverless runtime for request handling
• Cloudflare Workers AI: LLM inference using Llama 3.1 8B Instruct Fast
• Durable Objects: Persistent state and session management
• TypeScript: Type-safe development
• Vitest: Testing framework with Cloudflare Workers support

Additional Resources

• Cloudflare Workers Documentation
• Cloudflare Workers AI Documentation
• Durable Objects Documentation
• Cloudflare Pages Documentation