🧠 2USE - The Doubled Understanding AI

"Dua kali lipat yang dapat kamu lihat" - Melipatgandakan kemampuan berpikir, bukan ketergantungan

2USE (Double U-S, dibaca "double-use" atau "the doubled that you can see") adalah custom AI middleware yang menolak memberikan jawaban instan dan melatih pengguna untuk berpikir kritis melalui metode Socratic.

Dikembangkan sebagai core intelligence untuk We Will Shine - aplikasi produktivitas belajar dengan UI stoik.

---

🎯 What is 2USE?

The Problem:

Traditional AI (ChatGPT, Gemini, etc.):
User: "Solve 2x + 3 = 7"
AI: "x = 2. Here's the solution: 2x + 3 = 7, subtract 3..."
└─ User gets answer but learns nothing ❌

Result: AI dependency, no critical thinking, passive learning

The 2USE Solution:

2USE AI:
User: "Solve 2x + 3 = 7"
2USE: "Great question! Before I give hints, tell me:
      1. What are you trying to find?
      2. What's 'blocking' the x? How can you remove it?
      3. What have you tried so far?
      
      Write your attempts, then we'll continue together! 💡"
└─ User forced to think, then guided step-by-step ✅

Result: Independent thinking, deeper understanding, active learning

The Philosophy: "The Doubled"

2USE = See process + See result = 2x understanding

Without 2USE (1x)	With 2USE (2x = "The Doubled")
See only answer	See journey + destination
Copy-paste	Think + Learn
Instant gratification	Lasting comprehension
AI dependency	AI-assisted independence

---

✨ Features

Core Capabilities:

• 🚫 Anti-Instant Answer - Refuses direct solutions, forces thinking
• 🤔 Socratic Method - Asks guiding questions, not giving answers
• 📚 Context-Aware - Integrates with notes, tasks, study sessions
• 🎯 Progressive Hints - Level 1 → 2 → 3 → 4 (gradually more helpful)
• ⏱️ Thinking Time Enforcement - Requires 30s+ of thinking before answering
• 📊 Dependency Monitoring - Tracks usage patterns, prevents addiction
• 🛡️ Homework Detection - Blocks cheating attempts, promotes learning
• 🧩 Pattern Caching - 70% cost reduction through Socratic response caching

Technical Features:

• Multi-LLM Support - Gemini, OpenAI, Claude, Ollama
• Middleware Architecture - Wraps existing LLMs with custom behavior
• Cost Optimized - Free tier maximization, intelligent caching
• Privacy First - Local-first option, no data selling
• Open Source - MIT License, community-driven

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                        User Input                           │
│              "Gimana cara solve 2x + 3 = 7?"                │
└──────────────────────┬──────────────────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────────────────┐
│                   2USE Middleware Layer                      │
├─────────────────────────────────────────────────────────────┤
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐     │
│  │   Intent     │  │   Context    │  │   Thinking   │     │
│  │  Classifier  │  │   Analyzer   │  │Time Checker  │     │
│  └──────────────┘  └──────────────┘  └──────────────┘     │
│                                                              │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐     │
│  │  Dependency  │  │   Socratic   │  │   Pattern    │     │
│  │   Monitor    │  │Prompt Engine │  │    Cache     │     │
│  └──────────────┘  └──────────────┘  └──────────────┘     │
└──────────────────────┬──────────────────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────────────────┐
│              Base LLM (Gemini/GPT/Claude/Ollama)            │
│                  "Enhanced System Prompt"                   │
└──────────────────────┬──────────────────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────────────────┐
│                2USE Response Modifier                        │
├─────────────────────────────────────────────────────────────┤
│  • Remove any leaked direct answers                         │
│  • Insert guiding questions                                 │
│  • Add reflection prompts                                   │
│  • Select appropriate hint level                            │
└──────────────────────┬──────────────────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────────────────┐
│                   Socratic Response                          │
│    "Oke! Sebelum aku kasih hint, coba jawab dulu:          │
│     1. Apa yang kamu mau cari?                              │
│     2. Angka 3 itu 'ngeganggu' ga? Gimana atasinnya?        │
│     Tulis jawabanmu dulu, nanti kita lanjut!"               │
└─────────────────────────────────────────────────────────────┘

---

🚀 Quick Start

Installation

npm install 2use-ai
# or
yarn add 2use-ai
# or
pnpm add 2use-ai

Basic Usage

import { TwoUse } from '2use-ai';

// Initialize 2USE
const twoUse = new TwoUse({
  provider: 'gemini', // or 'openai', 'claude', 'ollama'
  apiKey: process.env.GEMINI_API_KEY, // optional for free tier
  options: {
    enableThinkingTimeEnforcement: true,
    minimumThinkingTime: 30000, // 30 seconds
    maxDailyInteractions: 50,
    enableHomeworkDetection: true,
    cachingEnabled: true
  }
});

// Ask a question
const response = await twoUse.ask({
  question: "Berapa hasil dari 2x + 3 = 7?",
  context: {
    currentNote: "Studying algebra chapter 3",
    previousAttempts: 0,
    userLevel: "beginner"
  }
});

console.log(response.message);
// "Oke! Sebelum aku kasih hint, coba jawab dulu..."

With Context Integration

// Example: Integration with note-taking app
const twoUse = new TwoUse({ provider: 'gemini' });

const response = await twoUse.ask({
  question: "Explain photosynthesis",
  context: {
    currentNote: {
      title: "Biology - Photosynthesis",
      content: "Plants make food using...",
      tags: ["biology", "plants"]
    },
    studySession: {
      subject: "Biology",
      startTime: Date.now() - 1800000, // 30 minutes ago
      tasksCompleted: 2
    },
    userHistory: {
      previousQuestions: ["What is chlorophyll?"],
      conceptsMastered: ["cell structure", "respiration"]
    }
  }
});

---

📖 API Reference

TwoUse Class

Constructor

new TwoUse(config: TwoUseConfig)

Config Options:

interface TwoUseConfig {
  provider: 'gemini' | 'openai' | 'claude' | 'ollama';
  apiKey?: string;
  options?: {
    enableThinkingTimeEnforcement?: boolean;
    minimumThinkingTime?: number; // milliseconds
    maxDailyInteractions?: number;
    enableHomeworkDetection?: boolean;
    enableLazyPromptDetection?: boolean;
    cachingEnabled?: boolean;
    progressiveHintLevels?: number;
  };
}

Methods

ask(params: AskParams): Promise

interface AskParams {
  question: string;
  context?: {
    currentNote?: any;
    currentTask?: any;
    studySession?: any;
    userHistory?: any;
  };
  userId?: string;
}

interface TwoUseResponse {
  message: string;
  hintLevel: 0 | 1 | 2 | 3 | 4;
  requiresThinkingTime: boolean;
  detectedIntent: 'genuine' | 'lazy' | 'homework' | 'copied';
  suggestions?: string[];
  reflectionPrompts?: string[];
}

checkDependency(userId: string): Promise

interface DependencyScore {
  score: number; // 0-100, lower is better
  status: 'healthy' | 'warning' | 'concerning' | 'critical';
  dailyUsage: number;
  independenceRate: number; // % of problems solved without AI
  recommendation: string;
}

---

🎨 Response Modes

2USE operates in different modes based on user intent:

1. Socratic Challenge (Default)

User: "What is photosynthesis?"

2USE Response:
"Before I explain, let's explore together:
1. Why are leaves green, not red?
2. What happens if plants are kept in darkness?
3. Where do plants get their energy?

Answer these first, then we'll dive deeper!"

2. Homework Blocker

User: "Do my math homework questions 1-10, deadline tomorrow"

2USE Response:
"I noticed this looks like homework 📚
I won't give direct answers, but I can help you UNDERSTAND!

Let's start with #1:
- Try it yourself first
- Show me your step-by-step thinking
- I'll point out where you're stuck
- You'll learn and be ready for tests!"

3. Lazy Prompt Detector

User: "answer this" *paste question*

2USE Response:
"Hmm, seems like you haven't tried thinking first 😅

Tell me:
1. What do you understand from the question?
2. What confuses you?
3. What have you tried?

Just 2-3 sentences is enough!
Then I can give hints that actually help 🧠"

4. Progressive Hints

// Attempt 1 (Level 0 - Socratic only)
"What are you trying to find? What's blocking you?"

// Attempt 2 (Level 1 - High-level direction)
"Think about isolating the variable. What operation would help?"

// Attempt 3 (Level 2 - Break down the problem)
"The +3 is on the wrong side. What's the opposite of +3?"

// Attempt 4 (Level 3 - Similar example)
"Example: If x + 5 = 10, we subtract 5 from both sides..."

// Attempt 5 (Level 4 - Step-by-step, last resort)
"Okay, let's solve together: 2x + 3 = 7
 Step 1: Subtract 3 from both sides..."

---

⚙️ Configuration Examples

Free Tier Optimization

const twoUse = new TwoUse({
  provider: 'gemini', // 1500 free requests/day
  options: {
    cachingEnabled: true, // Reduce API calls by 70%
    progressiveHintLevels: 4
  }
});

Local Privacy-First

const twoUse = new TwoUse({
  provider: 'ollama', // Runs locally
  options: {
    model: 'llama3.1', // or mistral, phi
    cachingEnabled: true
  }
});

Hybrid Setup (Best of Both Worlds)

const twoUse = new TwoUse({
  providers: [
    { name: 'ollama', priority: 1, use: 'simple-questions' },
    { name: 'gemini', priority: 2, use: 'complex-reasoning' }
  ],
  options: {
    autoRouting: true // Automatically choose provider
  }
});

---

📊 Cost Estimation

Setup	Monthly Users	Est. Cost	Notes
Free Tier	1,000	$0	Gemini Flash + aggressive caching
Hybrid	5,000	$15-30	Ollama (local) + Gemini (complex)
Scale	10,000+	$50-100	Multi-provider with fallbacks

Cost Breakdown (1000 active students):

• 50% use AI daily = 500 users
• Avg 10 messages/day = 5,000 messages
• 70% cached = 1,500 LLM calls
• Gemini Flash free tier: $0 ✅
• GPT-4o mini fallback: ~$13.50/month

---

🛡️ Safety Features

Anti-Cheating Mechanisms

// Homework pattern detection
const homeworkPatterns = [
  /kerjain PR/i,
  /tugas \d+ nomor/i,
  /deadline (besok|tomorrow)/i,
  /jawaban langsung/i
];

// Copy-paste detection
const isCopied = detectCopyPaste(userMessage);

// Lazy prompt detection
const thinkingTime = now() - lastMessageTime;
if (thinkingTime < 30000) {
  throw new ThinkingTimeRequired();
}

Dependency Prevention

// Usage limits
const limits = {
  maxDailyMessages: 50,
  maxConsecutiveQuickQuestions: 5,
  minimumBreakTime: 600000 // 10 minutes
};

// Independence scoring
const independenceRate = (solvedAlone / totalProblems) * 100;
// Target: 70%+ (user should solve most alone)

---

🧪 Testing

# Run tests
npm test

# Test with different providers
npm test -- --provider=gemini
npm test -- --provider=ollama

# Test specific scenarios
npm test -- --scenario=homework-detection
npm test -- --scenario=lazy-prompt
npm test -- --scenario=progressive-hints

---

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for details.

Development Setup

# Clone repo
git clone https://github.com/sandikodev/2use.git
cd 2use

# Install dependencies
pnpm install

# Run in dev mode
pnpm dev

# Build
pnpm build

Areas We Need Help:

• 🌍 Multi-language support (currently ID/EN)
• 🧠 More Socratic prompt templates
• 🎯 Subject-specific hint strategies (Math, Science, etc.)
• 📊 Analytics dashboard
• 🔬 Research on learning effectiveness

---

📚 Research & Impact

Metrics We Track:

• Thinking Depth Score - How deeply users think before asking
• Self-Sufficiency Rate - % of problems solved independently
• Dependency Index - Early warning system for AI addiction
• Learning Progress - Actual comprehension vs. answer collection

Early Results (Beta, n=22 students):

• ✅ 68% self-sufficiency rate (target: 70%)
• ✅ 45% reduction in AI messages over 4 weeks
• ✅ 85% student satisfaction (despite initial frustration)
• ✅ Teacher-reported improvement in critical thinking

---

🗺️ Roadmap

v1.0 (Current)

• ✅ Socratic prompting engine
• ✅ Multi-LLM support
• ✅ Basic dependency prevention
• ✅ Homework detection
• ✅ Progressive hints

v1.1 (Q1 2026)

• Voice interface
• Image/diagram analysis
• Multi-language (10+ languages)
• Advanced caching strategies

v2.0 (Q2 2026)

• Fine-tuned model for education
• Predictive intervention (before dependency forms)
• Subject-specific expert modes
• Teacher analytics dashboard

v3.0 (Q3 2026+)

• Research API for educational institutions
• White-label licensing
• Integration marketplace
• Global impact measurement

---

📄 License

MIT License - see LICENSE for details.

---

🙏 Acknowledgments

2USE is developed by @sandikodev as part of We Will Shine project.

Inspired by:

• Socratic Method (ancient Greek philosophy)
• Growth Mindset (Carol Dweck)
• Zone of Proximal Development (Lev Vygotsky)
• Islamic educational principles
• Indonesian education context

Special Thanks:

• 22 siswa 7D Putri SMP IT Masjid Syuhada (beta testers)
• Educational psychology researchers
• Open source community

---

📞 Contact & Support

• GitHub Issues: Report bugs or request features
• Discussions: Join the community
• Email: 2use@wewillshine.dev
• Twitter: @wewillshine

---

💡 Fun Facts

Why "2USE"?

1. Akronim: Double U-S from We Will Shine
2. Pronunciations:
   • "Double-use" (how it sounds)
   • "Double you can see" (what it means)
   • "The doubled that you can see" (philosophical depth)
3. Bahasa Indonesia: "Dua kali lipat yang dapat kamu lihat"
4. Triple Meaning:
   • Doubled Understanding - See process + result
   • Two-way Interaction - Socratic dialogue
   • To Use (wisely) - Use AI without dependency

---

Built with ❤️ for independent thinkers

"The doubled that makes you unstoppable"

⭐ Star us on GitHub | 🚀 Try We Will Shine