# PATAS Overview

**Pattern-Adaptive Anti-Spam System** - A self-learning system that discovers spam patterns and generates blocking rules.

---

## What Problem Does PATAS Solve?

Platforms with user-generated content face a constant challenge: **spam evolves faster than manual rules can keep up**.

Traditional approaches:
- **Manual rule writing** - Slow, doesn't scale, misses new patterns
- **Static ML models** - Require retraining, may miss edge cases
- **Reactive blocking** - Always one step behind attackers

PATAS provides a **proactive, adaptive solution**:
- Automatically discovers new spam patterns from your data
- Generates blocking rules that can be deployed immediately
- Continuously learns and adapts as spam evolves

---

## How PATAS Works

### 1. Data Ingestion
PATAS ingests historical message data (spam and non-spam examples) from your platform.

### 2. Pattern Discovery
The system analyzes messages to identify recurring patterns:
- URLs and domains
- Phone numbers
- Keywords and phrases
- Message structure and signatures
- Language patterns

### 3. Rule Generation
Discovered patterns are converted into machine-readable blocking rules:
- SQL expressions for database filtering
- Rule definitions for rule engines
- Configurable precision and coverage

### 4. Safe Evaluation
Rules are tested in "shadow mode" before deployment:
- Applied to recent traffic without blocking
- Metrics collected: precision, recall, coverage
- False positive risk assessed

### 5. Deployment
High-quality rules are promoted to active status and can be exported for deployment to your filtering system.

---

## Key Features

### 🎯 Pattern Discovery
- Automatically identifies spam patterns from your data
- Supports multiple pattern types (URLs, keywords, signatures, etc.)
- Uses LLM for intelligent pattern recognition (optional)

### 🔒 Safe Rule Lifecycle
- **Candidate** → **Shadow** → **Active** → **Deprecated**
- Shadow evaluation prevents false positives
- Automatic rollback for degrading rules

### 📊 Metrics & Evaluation
- Precision, recall, coverage tracking
- False positive monitoring
- Performance metrics per rule

### 🚀 Production-Ready
- RESTful API for integration
- Batch processing for large datasets
- Configurable aggressiveness profiles (conservative/balanced/aggressive)

---

## Typical Use Cases

### Messaging Platforms
Platforms with user-to-user messaging need to block spam while avoiding false positives that frustrate legitimate users.

**How PATAS helps:**
- Discovers new spam patterns as they emerge
- Generates rules that can be deployed immediately
- Monitors rule performance and auto-deprecates bad rules

### Content Moderation Teams
Teams managing user-generated content need to scale their moderation efforts without hiring more moderators.

**How PATAS helps:**
- Reduces manual review workload
- Identifies patterns that humans might miss
- Provides explainable rules (not a black box)

### Anti-Spam Systems
Existing anti-spam systems need to adapt to new attack patterns without constant manual intervention.

**How PATAS helps:**
- Complements existing rules with discovered patterns
- Provides a continuous learning loop
- Integrates via rule export (SQL, JSON, etc.)

### Platform Operators
Platforms with growing user bases need automated spam detection that scales with traffic.

**How PATAS helps:**
- Handles large datasets efficiently
- Processes messages in batches
- Provides API for integration into existing infrastructure

---

## What PATAS Detects

PATAS focuses on **commercial spam** patterns:

✅ **Detected:**
- Buy/sell offers
- Job solicitations
- Commercial promotions
- Service advertisements
- Phishing attempts
- Suspicious URLs and domains

❌ **Out of Scope:**
- Political content
- Hate speech
- General toxicity
- Content moderation (beyond spam)

---

## Architecture Overview

```
Your Platform → PATAS API → Pattern Mining → Rule Generation → Rule Export → Your Filtering System
                     ↓
              Shadow Evaluation
                     ↓
              Metrics & Monitoring
```

**Key Components:**
- **API Layer** - RESTful endpoints for integration
- **Pattern Mining** - Discovers patterns from messages
- **Rule Lifecycle** - Manages rule states and transitions
- **Shadow Evaluation** - Tests rules safely before deployment
- **Rule Backend** - Exports rules in various formats

---

## Getting Started

1. **Run the Demo** - See [Demo Guide](DEMO_GUIDE.md) for a quick walkthrough
2. **Try the API** - See [API Quickstart](API_QUICKSTART.md) for integration examples
3. **Explore Use Cases** - See [Use Cases](USE_CASES.md) for real-world scenarios

---

## Next Steps

- [Demo Guide](DEMO_GUIDE.md) - Run a local demo
- [API Quickstart](API_QUICKSTART.md) - Integrate PATAS into your system
- [Use Cases](USE_CASES.md) - See how others use PATAS

---

**Note**: PATAS is a **pattern discovery and rule generation system**, not a real-time filter. It analyzes historical data and generates rules that you deploy to your filtering system.