DEV Community

Миша Ефремов
Миша Ефремов

Posted on

I Built an Autonomous AI SIEM With 10 Neural Networks

What if your server could defend itself?

That's the question that drove me to build SHARD — a fully autonomous cybersecurity system that detects attacks, generates real-time defense rules, blocks hackers, and predicts their next move. All without a security team. All without human intervention.

The Problem

Every day, thousands of servers are attacked. SQL injections, brute force attempts, DDoS floods, ransomware. Small businesses can't afford enterprise SIEM solutions like Splunk or Palo Alto ($50,000+/year). They need something that just works — automatically.

I decided to build it.

What SHARD Does

When an attacker hits your server:

  1. 13 honeypots detect the connection (SSH, MySQL, Redis, MongoDB, FTP, etc.)
  2. XGBoost ML model classifies the attack type (13 types, 100% accuracy)
  3. Seq2Seq Transformer (5.35M parameters) generates unique iptables/WAF rules
  4. RL DQN Agent decides: block permanently? block temporarily? throttle?
  5. VAE Anomaly Detector checks if this is a zero-day attack
  6. GNN Threat Graph maps the attacker's connections
  7. Temporal GNN predicts their next target
  8. Multi-Modal Fusion combines all 8 signals into one threat score
  9. Telegram/Slack notification is sent immediately
  10. Everything is logged and the attacker is blocked

The Numbers

Metric Value
Neural Networks 10
Attack Classification Accuracy 100%
RL Decision Accuracy 100%
Anomaly Detection Rate 91.2%
Defense Actions (1 hour test) 4,000+
RL Decisions (1 hour test) 8,000+
Throughput 870 packets/sec
Honeypots 13

Tech Stack

  • PyTorch for deep learning models
  • XGBoost for attack classification
  • Docker for one-command deployment
  • Swagger for API documentation
  • pytest for testing (11/11 passing)
  • Telegram API for notifications

Try It Yourself


bash
git clone https://github.com/misha622/shard-siem
cd shard-siem
docker build -t shard-siem .
docker run -d --name shard -p 8080:8080 -p 5001:5001 shard-siem
Enter fullscreen mode Exit fullscreen mode

Top comments (0)