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 — Multi-Label MLP
// 20 simultaneous attack pattern predictions
Input: [50-dim unified feature vector]
│
▼
Layer 1: Dense(50 → 128) // Leaky ReLU α=0.01 · Adam
│
▼
Layer 2: Dense(128 → 64) // Leaky ReLU · feature mixing
│
▼
Layer 3: Dense(64 → 32) // Leaky ReLU · compression
│
▼
Output: Dense(32 → 20) // Sigmoid per pattern
Loss: Focal(γ=2) + POS_WEIGHT=8
// corrects for ~5% positive rate
// per output class
Optim: Adam β1=0.9 β2=0.999 ε=1e-8
Threshold: 0.50 // binary classification cutSelect 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 VAPT finding, dark web alert, console threat, red team engagement, and brand monitoring hit is captured as a training signal — no manual labeling needed.
The training engine maps each event type to the relevant attack patterns using deterministic rules. A credential dump maps to credential_stuffing + credential_dump_ntlm. An XSS event maps to xss_session_hijack.
A single gradient step is applied to the entity's model weights immediately — no batch wait, no retraining pipeline. The model learns from each new finding within milliseconds.
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.