
1
1
Slovakia
2+ years of experience
CS Student at STU Bratislava. Passionate about AI engineering and Cybersecurity. I prefer diving deep into model weights rather than just prompting. Currently building Fenrir — a specialized fine-tuned LLM for automated vulnerability detection and secure code analysis

The Problem: Modern cybersecurity operations are overwhelmed by complex, obfuscated scripts. Traditional static analysis fails to capture the true intent of an attacker, forcing SOC analysts into slow, manual de-obfuscation tasks that delay incident response.The Solution: DeepSeek-Cybersec-33B is a specialized forensic engine designed to automate threat intelligence. By fine-tuning the DeepSeek-33B architecture using LoRA on AMD Instinct™ MI300X accelerators, we created a model capable of understanding code "intent" beyond mere syntax. It identifies malicious patterns like Discord webhooks, data exfiltration, and credential theft , providing structured forensic verdicts: CLEAN, SUSPICIOUS, or MALICIOUS. Target Audience & Impact: Designed for SOC teams and forensic investigators, our engine reduces initial triage time by up to 80%. With over 13 community downloads on Hugging Face before official release, it demonstrates strong immediate demand. Integrated with a Streamlit interface, it offers a seamless, high-performance workflow for deep code audits.
10 May 2026