The CtrlSec Brain is a multi-label MLP trained exclusively on your organisation's security events. 20 simultaneous attack pattern predictions. Zero external dependencies. Zero cloud LLM latency.
// CtrlSec Brain — 9 Self-Learning Engines
// 2M+ training signals · 20 attack patterns · 50-dim features
Engine 1: AttackClassifier MLP (50→128→64→32→20)
Focal-BCE loss · AdaGrad optimizer · 20 sigmoid outputs
Engine 2: AttackPredictor (5 independent MLPs, 20→16→8→1)
Ransomware · Data Breach · Supply Chain · Credential · APT
Engine 3: VAPT ReinforcementLearner
Genetic algorithm · Tournament selection · Payload evolution
Engine 4: ThreatNeuralScorer (15→32→16→1)
Online learning · Per-entity weights · <100ms inference
Engine 5: AnomalyDetector Autoencoder (12→6→4→6→12)
Reconstruction error scoring · Adaptive thresholds
Engine 6: SequenceAnalyzer (Kill chain state machine)
NORMAL → SUSPICIOUS → ESCALATING → ACTIVE_KILL_CHAIN
Engine 7: FindingValidator (5-check true-positive gate)
Confidence · Duplicate · Neural · CVE/CWE · Severity
Engine 8: CrossModuleFusion (10 correlation pairs, 2.5x weight)
Multi-source intelligence amplification
Engine 9: SelfLearner (7-phase continuous cycle, 6hr cron)
HARVEST → CORRELATE → TRAIN → EVOLVE → PREDICT → ASSESS → REPORTSelect a pre-built attack scenario below. The request hits the production model with hardcoded 50-dim vectors — no real client data is used. Results reflect the current trained weights.
Every 6 hours, the self-learning engine harvests signals from all 7 security modules — VAPT, Red Team, Dark Web, Supply Chain, Brand, GRC, and Console threats. Each signal becomes a training event.
Each event is auto-labeled to 20 attack patterns using deterministic rules. A 5-check validation gate (confidence, duplicate, neural, CVE/CWE, severity) filters true positives from noise before any finding reaches the database.
Signals confirmed by multiple engines get 2.5x weight amplification. The MLP classifier trains via curriculum learning (easy→expert), self-play reinforcement (predict→correct), mutation evolution, and adversarial hardening.
VAPT payloads evolve via genetic algorithm. 5 independent MLPs update attack forecasts. The strategy orchestrator re-scores overall risk posture across all companies. Admin reviews and approves findings before clients see them.
Nine interconnected engines continuously ingest security signals, cross-correlate across modules, and self-improve through reinforcement learning, genetic algorithms, and multi-source intelligence fusion. No external APIs. No cloud dependencies. Your data never leaves your infrastructure.
50→128→64→32→20 multi-label neural network
20 attack patterns · Focal-BCE · AdaGrad
5 independent MLPs forecasting attack probability
Ransomware · Data Breach · Supply Chain · Credential · APT
Genetic algorithm evolving 10,000+ payloads
Tournament selection · Mutation · Crossbreeding
150+ MITRE ATT&CK TTPs · 20 adversary profiles
41 TTPs tracked · Kill chain simulation
Vendor risk · CVE correlation · SBOM analysis
9 records · 80+ CVEs · License risk
Typosquatting · Phishing · Impersonation detection
31 patterns · 12 mutation algorithms
6 compliance frameworks · 140+ controls
19 controls · SOC2 · ISO27001 · NIST · GDPR · PCI · HIPAA
Multi-layer neural network + anomaly autoencoder
Online learning · Kill chain sequence analysis
Multi-source intelligence correlation at 2.5x weight
10 correlation pairs · Temporal decay · Adversarial training
The Brain trains on your data — not a shared cloud model. Every VAPT finding, dark web alert, and red team event makes your model smarter. Fully isolated. Fully yours.