Research
Paper accepted at IEEE WETICE 2026
AI-driven threat knowledge platform inspired by MITRE ATT&CK.
Read the abstract// AI offensive security · LLM red-teaming
AI offensive security, Red-teaming LLMs @ AnTitech, CTF player · CSAW '26 finalist, Challenge author, Cybersecurity Master's student
Cybersecurity Master's student and CTF player. I break LLMs and chatbots before attackers do — jailbreaks, prompt injection, data poisoning — then write up how.
// 00 — signal
Research
AI-driven threat knowledge platform inspired by MITRE ATT&CK.
Read the abstractCTF
1st in MENA, 2nd worldwide. Qualified for the Finals.
Read the CSAW write-ups// 01 — operator profile

Master's student in Cybersecurity with a strong background in software engineering, CTF competitions and Python automation. Excellence Scholarship recipient at MedTech University, and a full-scholarship alum of the University of Zurich's Deep Dive into Blockchain summer school.
Red-teaming LLMs and chatbots — jailbreaks, prompt injection and data poisoning — and building the tooling that helps autonomous agents find new ones.
Web, forensics, MISC and pyjails. Competitor, team lead and challenge author for national CTFs.
Dark Web intelligence mapped to MITRE ATT&CK, OSINT automation and vulnerability operations.
// 02 — field notes
CTF writeups & security insights
Three chained bugs and a deliberate canary. The obvious path ends on a planted flag that gets you banned; the real one needs a client-side attempt counter and a 10,000-PIN sweep.
Two first-party identity providers, one of which forgot to pin its signing certificate. An error oracle in the callback exposed the gap, and the admin gate keyed off scope rather than sub.
The per-character delay multiplier came from your own User-Agent. Crank it up and the signal saturates — most of the work was stopping the measurement methodology from lying.
// 03 — arsenal
Tools, AI platforms & security tooling
Final project for Group 13 at the University of Zurich's Deep Dive into Blockchain summer school. Closes the trust gap that keeps ocean cleanup underfunded: a marketplace where verified collectors are paid directly in OCT tokens for proven waste retrieval. Tamper-resistant payout model verified independently before release and weighted by material value and retrieval difficulty, plus a self-sustaining funding loop routing recycler purchases and donor contributions into a liquidity pool.
No public repository
Multi-source founder data aggregation engine (FastAPI, aiohttp, asyncio) normalizing LinkedIn, GitHub, Product Hunt and DEV.to profiles into structured JSON. Identity resolution via multi-engine search cascade + GitHub API verification. LightGBM classifier predicting startup acquisition/IPO success — ROC-AUC 0.84 on 196K companies.
Privacy-first mobile + backend system that scans documents, images, audio and calendars on-device and indexes them with a local LLM (Qwen2.5-3B via Ollama) — no data ever leaves the network. Multi-modal ingestion (PyMuPDF, BLIP, Whisper), self-verifying RAG pipeline with ChromaDB, React Native/Expo app. Built for AIMinds Hackathon (5th place).
Research project accepted at IEEE WETICE 2026 (Paris). Aggregates Dark Web intelligence and maps threat actors, techniques and indicators to the MITRE ATT&CK framework to proactively enrich threat knowledge bases.
No public repository