Security Researcher | Self-Taught Developer | Local LLM & Autonomous Agent Builder
Building Halo — an autonomous penetration testing agent powered by local models. I work at the intersection of security, AI, and infrastructure.
I'm obsessed with:
- 🔐 Autonomous Security Agents — orchestrating pentest workflows with local LLMs (Gemma 4 12B Abliterated via LM Studio)
- 🧠 Local AI Deployment — optimizing models for on-device inference (no cloud, no API keys)
- 🛠️ Security Tooling — building agents that coordinate 22 integrated security tools with intelligent recon-to-exploit workflows
- 📱 Mobile Security — porting models to iOS/Android for offline vulnerability assessment
- 🔄 Agent Infrastructure — memory systems, logging, negative experience caching, decision tracking
- Recon: masscan, nmap, nikto
- Exploitation: sqlmap, hydra, ncrack, searchsploit
- Payloads: msfconsole, social engineering toolkit
- Analysis: Burp Suite, metasploit, custom agents
Local-Model — Halo v0.1.0
Autonomous Penetration Testing Agent
An intelligent, autonomous agent for coordinated penetration testing. Runs Gemma 4 12B Abliterated locally via LM Studio, orchestrates 22 integrated security tools via MCP server, maintains persistent memory, learns from negative experiences, and generates comprehensive pentest reports.
Key Features:
- ✅ Autonomous tool orchestration — intelligent recon-to-exploit workflows (masscan → nmap → exploit chains)
- ✅ Persistent negative experience cache — prevents redundant attempts, optimizes testing
- ✅ JSONL-backed short-term memory — decision tracking, learning-based strategy refinement
- ✅ HTML pentest report generation — professional, actionable findings
- ✅ 22 integrated security tools — comprehensive attack surface coverage
- ✅ Tested against Metasploitable — 14/24 ports successfully breached
Tech Stack: Python, Flask, Gemma 4 12B Abliterated, LM Studio, MCP Server, Kali Linux
Status: v0.1.0 Released | Production-Ready | Actively Maintained
GEMMA-by-GOOGLE — Enhanced Security Framework
Gemma 4 12B Cybersecurity Implementation
High-performance autonomous security agent leveraging Google's Gemma 4 12B (abliterated) model. Deployed on Linux with persistent learning capabilities, automatic report generation, and full local deployment.
Key Features:
- ✅ Fully local deployment — no cloud dependencies
- ✅ Persistent negative experience cache — intelligent learning optimization
- ✅ Autonomous recon-attack loops — 22 integrated security tools
- ✅ Professional report generation — actionable security findings
- ✅ Learning optimization — improves with each engagement
Tech Stack: Python, Gemma 4 12B, Linux environment, LM Studio
Qwen-2.5-1.5B-Android ✅ Published
Mobile Security Model for Android
Quantized Qwen 2.5-1.5B model optimized for Android devices. Enables on-device inference for security assessments without cloud dependency.
What's Inside:
- Multiple quantization formats (Q8 GGUF, Q4 GGUF, SafeTensors)
- Tested on Android 12+
- Low-latency inference
- Privacy-first (all processing local)
Use Cases: Security researchers, penetration testers, red teamers needing portable threat modeling
Qwen-2.5-1.5B-iOS ✅ Published
Mobile Security Model for iOS
Quantized Qwen 2.5-1.5B uncensored model for iOS devices. Run security assessments directly on iPhone/iPad without internet connectivity.
What's Inside:
- Q8 GGUF, Q4 GGUF, SafeTensors formats
- iOS 15+ compatible
- Optimized for Off Grid app + local ML frameworks
- 341+ downloads in first 24 hours
Use Cases: On-device vulnerability analysis, offline threat assessment, edge security workflows
| Project | Status | Model | Tools | Release |
|---|---|---|---|---|
| Halo | Active | Gemma 4 12B | 22 | v0.1.0 ✅ |
| GEMMA-by-GOOGLE | Active | Gemma 4 12B | 22 | Published |
| Qwen Android | Published | Qwen 2.5-1.5B | Mobile | Live |
| Qwen iOS | Published | Qwen 2.5-1.5B | Mobile | Live |
I publish quantized models, Spaces, and security datasets on Hugging Face: automajicly
Published Models & Resources:
- ✅ Qwen 2.5-1.5B Android — Quantized variants (Q8 GGUF, Q4 GGUF, SafeTensors) — Live & Production-Ready
- ✅ Qwen 2.5-1.5B iOS — Uncensored quantized model — Live & Production-Ready
- 🚀 Halo Security Agent — v0.1.0 with Gemma 4 12B
- 📱 iOS/Android Spaces — Interactive Gradio demos & documentation
I'm actively looking for collaborators on:
- Autonomous agent improvements — enhanced tool selection algorithms, exploit chain optimization
- Model optimization — advanced quantization techniques, inference speed improvements
- Security tooling — novel exploits, recon modules, detection evasion research
- Mobile deployment — iOS/Android frameworks, edge inference optimization
- Open-source security infrastructure — community-driven tool development
Interested in collaborating? Let's connect below.
- Email: christopherwsheridan@gmail.com
- GitHub: @XenoCoreGiger31
- Hugging Face: @automajicly
- GitHub Sponsors: Support the work
- 🐍 Python Mastery — deep systems-level Python, performance optimization, architectural patterns
- 💾 Systems & Infrastructure — OS internals, networking, memory management, distributed systems
- 🔐 Advanced Security Research — rigorous vulnerability analysis, exploit development, threat modeling
- 🛠️ Production Tool Development — building enterprise-grade security infrastructure
- 📖 Technical Leadership — documentation, mentoring, transparent knowledge sharing
I'm self-taught and learn by building real tools to solve real problems:
- Public Learning — all work on GitHub with documented reasoning and transparent decision-making
- Deep Expertise — mastery from fundamentals to production-grade implementation
- Security-First — rigorous testing, transparent security findings, responsible disclosure
- Quality Infrastructure — reliable, reusable, maintainable security tools
- Knowledge Sharing — collaborative development, no gatekeeping, community-driven growth
- Continuous Improvement — iterating on ideas, embracing feedback, learning from failures
Guiding Principle: Build tools that work, document why they work, share knowledge freely.
Last updated: June 2026 | Always building | No AI-generated BS 🚀