Automates incident response, playbooks, and enterprise workflows.
Analyzes logs, emails, and user behavior to uncover threats.
Safeguards research, designs, and sensitive data across all devices.
ARRTECH's threat detection engine is built on a class of brain-inspired machine learning called predictive error-compensated neural networks — a method that mirrors how the brain reduces uncertainty by forecasting its own errors and correcting in real time.
The PEC-WNN architecture was introduced in peer-reviewed research published in IEEE Access, 2020, by Prof. Dr. Burak Berk Üstündağ of Istanbul Technical University.1 The work was conducted under the named research project Development of Cyberdroid Based on Cognitive Intelligent System Applications, funded through ITUNOVA, ITU's official technology transfer office. The published model reduced prediction error by up to 95% against prior art across four independent benchmark problems.
In benchmark testing against twenty leading time-series prediction methods, PEC-WNN delivered the lowest error rates while using a fraction of the parameters of comparable deep learning approaches.1 This architecture is the academic foundation underneath CYBERDROID Neural Detection — ARRTECH's threat detection engine.
Each family targets a distinct class of adversary behavior. Findings from multiple families are fused into coherent attack-chain narratives via cross-instrument correlation. No single-method blind spots.
Five additional proprietary algorithm families — covering identity drift, kernel-level data egress, and cross-product attack-chain reconstruction — are not enumerated publicly. Available under NDA during technical evaluation.
Cortical coding and the 80+ algorithm portfolio are not laboratory curiosities. Two production surfaces deliver them to working SOC teams every day.
CYBERDROID is the production implementation of the cortical coding research. It ingests telemetry from firewall, SIEM, SOAR, EDR, NDR, IAM, DLP, WAF, IDS/IPS, email, and cloud workload protection — then fuses findings from across the 15 algorithm families into a single actor-centric reasoning graph.
.avif)
SOC Analyst is where the detection engine meets the human operator. Every significant event becomes a structured investigation with actor context, raw evidence chains, hypothesis-and-dissent reasoning, and recommended next actions — not a line in a queue waiting to be triaged.
ARRTECH is not assembled from acquired startups. The platform is the product of a single continuous research program — and the same people who designed it lead the company.
Educated in computer engineering at Istanbul Technical University. Two decades building enterprise cybersecurity platforms now deployed across defense, telecom, government, and financial sectors globally. Work spans SIEM, SOAR, DLP, and AI-driven threat detection. Under his leadership the underlying platform has scaled to 3,000+ enterprise deployments across more than 25 large-scale organizations.
One of the foremost researchers globally in brain-inspired AI, neuromorphic computing, and signal processing applied to anomaly detection. Full Professor in the Faculty of Computer and Informatics Engineering at Istanbul Technical University; Director of the National Software Certification Research Center. Senior author of the foundational cortical coding research underlying CYBERDROID Neural Detection.
0000-0001-8143-9434Numbers below are drawn from the underlying platform's deployment record across two decades. Each is independently verifiable on technical evaluation under NDA.
A continuous five-year research program in predictive error-compensated neural architectures — the academic foundation underneath CYBERDROID Neural Detection. The full bibliography, including unpublished proprietary algorithm research, is available under NDA during technical evaluation.
High-Performance Time Series Prediction With Predictive Error Compensated Wavelet Neural Networks.
IEEE Access, 8, 210532–210541 — IEEE. Funded research project: "Development of Cyberdroid Based on Cognitive Intelligent System Applications" (ITUNOVA, 2019–2020).
Stock Price Prediction Using Predictive Error Compensation Wavelet Neural Networks.
Computers, Materials & Continua, 68(3), 3577–3593. Demonstrates 33% RMSE improvement over LSTM in real-world financial time-series prediction.
Predictive Error Compensating Wavelet Neural Network Model for Multivariable Time Series Prediction.
TEM Journal, 10(4), 1955–1963. Extends the PEC-WNN architecture to multivariable input — the configuration applicable to multi-source security telemetry.
Improvement in Prediction Performance Using Predictive Error Compensated Neural Networks.
Springer Lecture Notes in Networks and Systems. Continuing research extending the PEC-WNN architecture; demonstrates ongoing program investment through 2024.
A 60-minute working session with the engineers who built the platform. We walk you through the architecture, the algorithm portfolio, a live investigation on your own telemetry, and the full research bibliography under NDA.
Under mutual NDA · Engineer-led session · No sales overlay
