I'm a cybersecurity engineering student who learns by building things end to end — offensive security and vulnerability research, defensive tooling and detection, applied ML, and the systems and infrastructure underneath all of it. I like problems that cross those boundaries: an encrypted C2 framework, a honeypot platform with an ML analysis backend, a fine-tuned model that reasons about ATT&CK, a browser you can drive over a socket.
Everything here is built for research and authorized testing. Most of it is open source and self-hosted. Open to cybersecurity internships and entry-level roles.
| Project | Description |
|---|---|
| wisp | Lightweight encrypted C2 framework. Go implant, Python server, X25519 forward secrecy. |
| tuffyrat-c2 | RAT framework for authorized red-team ops — Python C2 with web UI, C++17 Windows agent, HTTP/HTTPS transport. |
| Exim RCE | Remote code execution research against the Exim mail transfer agent. |
| WiFi Pentest Assistant | Menu-driven Bash tooling for authorized wireless assessments — monitor mode, handshake capture, post-exploitation. |
| Project | Description |
|---|---|
| honeypot-ui | HoneySentinel AI — full-stack honeypot platform: protocol emulators, an analysis backend that classifies and enriches each session, and a review dashboard. |
| Chimera-14B-v2 | Cybersecurity reasoning LoRA for Ministral 3 14B — ATT&CK-aware blue-team analysis. |
| Project | Description |
|---|---|
| aegis | Red-team and guardrail benchmark for tool-using LLM agents — measures how much a layered defense stack cuts prompt-injection attack success, and what it costs in utility. Offline and reproducible. |
| abliteration | Training-free refusal-ablation on hybrid-attention models that standard interpretability tools can't load. Refusal 78.1% → 9.4%, capability unchanged. |
| Chimera-14B | Dual-capability LoRA (tool calling + extended reasoning) for DeepSeek-R1-Distill-Qwen-14B. |
| amnis | Persistent memory, RAG, and wiki compilation for AI agents — MCP server plus a web UI with an interactive knowledge graph. |
| aether | Local-first AI desktop agent: 23 tools, ReAct loop, native function calling, four permission tiers. Fully offline. |
| Project | Description |
|---|---|
| browseye | Full browser control for Firefox over WebSocket — tab management, DOM manipulation, extraction, network monitoring, all from the terminal. |
| pi-media-pipeline | Documented Raspberry Pi 5 media pipeline build. |
| pi-hole-uae | Pi-hole setup plus a curated blocklist of UAE-specific ad domains, telecom trackers, and regional ad networks. |
| Arch-Optimization-Script | Post-install performance tuning for Arch Linux — pacman, mirrors, ZRAM, boot. |
- Languages: Python, Go, C, C++, C#, Bash, JavaScript/TypeScript, SQL
- Backend & data: FastAPI, Flask, Node.js, PostgreSQL, SQLite, Redis
- Infra & ops: Linux, Docker, Nginx, Cloudflare, Raspberry Pi, GitHub Actions
- Security: C2 development, exploit research, honeypots, wireless assessment, MITRE ATT&CK
- ML: LoRA fine-tuning, RAG, agent loops (ReAct / function calling), MCP
