An intelligent, privacy-first Instagram account analyzer powered by cutting-edge Generative AI and Local LLM technology. Uses sophisticated AI reasoning to detect spam, inappropriate content, and bot accounts while maintaining complete privacy through local processing.

• 🤖 Local AI Analysis - Ollama + LLaMA 3.2 with chain-of-thought reasoning
• 🔒 Privacy First - 100% local processing, zero cloud dependencies
• 🎯 Multi-Dimensional Scoring - Spam, content, bot, quality detection
• ⚡ Smart Caching - 90%+ cache hits, sub-second responses
• 📊 Batch Processing - Analyze hundreds of accounts efficiently
• 🔍 Transparent AI - See exactly how AI reaches each decision
• 📥 Export Results - Download analysis as CSV

# Setup and run
git clone <your-repo-url>
cd instasanity
make setup
make run

# Access at http://localhost:8501

Requirements: Docker, 4GB+ RAM, Optional GPU

┌─────────────────────────────────────────────────────────────┐
│                  🧠 LOCAL LLM LAYER                        │
├─────────────────┬──────────────────┬─────────────────────────┤
│   Streamlit UI  │  Analysis Core   │   🦙 Ollama Engine      │
│   • Real-time   │  • Multi-criteria│   • Chain-of-Thought    │
│   • Interactive │  • Batch Process │   • Content Moderation  │
└─────────────────┴──────────────────┴─────────────────────────┘
         │                │                        │
         ▼                ▼                        ▼
┌─────────────────┐ ┌──────────────┐ ┌─────────────────┐
│ Instagram API   │ │  Smart Cache │ │  Redis Session  │
│ (Data Source)   │ │  (24h TTL)   │ │  (State Mgmt)   │
└─────────────────┘ ┌──────────────┘ └─────────────────┘

# Multi-step reasoning for transparent decisions
analysis_prompt = """
Think step by step:
1. SPAM: promotional language, follow-for-follow patterns
2. CONTENT: inappropriate material, violence, hate speech  
3. BOTS: username patterns, behavioral signals
4. QUALITY: authenticity, completeness

Return JSON with reasoning, scores, confidence...
"""

• Smart Caching: MD5 cache keys, 24h TTL, 90%+ hit rates
• Batch Processing: 200+ accounts in <2 minutes with caching
• Fallback Intelligence: Rule-based backup when LLM unavailable
• Real-time Metrics: Cache efficiency, confidence scoring

• Local Processing: No external API calls, Ollama runs locally
• Data Isolation: All analysis happens on your machine
• Session Management: Secure Instagram API integration
• Zero Telemetry: No usage data sent anywhere

• 0-3: Keep following (good accounts)
• 4-6: Review manually (questionable accounts)
• 7-10: Consider unfollowing (problematic accounts)

Analysis includes: Overall score, detailed reasoning, confidence level, content flags, account metadata

make setup    # Install AI models
make run      # Start application  
make stop     # Stop services
make logs     # View logs
make clean    # Clean containers

1. InstaSanity (8501) - Main application
2. Ollama (11434) - AI model server
3. Redis (6379) - Caching layer

make dev-build && make dev-run  # Development mode
make logs-app                   # Application logs
make logs-ollama               # AI model logs

Key settings in config.py:

• Ollama model and host configuration
• Cache TTL and directory settings
• Instagram API rate limits

1. Fork repository
2. Create feature branch
3. Test with make dev-run
4. Submit pull request

MIT License. For educational/personal use only. Always review AI results manually before taking action. Respect Instagram's Terms of Service.