I build practical AI systems that survive real constraints: limited VRAM, free-tier APIs, Windows/Kali workflows, messy browser automation, and tools that need to keep working after the demo.
| Agent Infrastructure MCP servers, memory layers, tool routers, schedulers, fallback chains, and workspace automation. |
Local AI Systems Ollama, Qwen-style local models, ChromaDB cache, voice/vision utilities, and offline fallback design. |
Security Automation Authorized OSINT, recon pipelines, Nmap/Shodan/Nuclei workflows, browser testing, and CEH-aligned methodology. |
|
MAST is my main build: a local-first AI operator that connects LLM routing, MCP tools, memory, browser automation, voice/vision utilities, and defensive security workflows into one workspace. It is designed for consumer hardware, unstable free-tier APIs, and real daily work where the assistant needs context, tools, and recovery paths instead of just chat. |
Hardware target: RTX 2060 Super, 8GB VRAM, local fallback ready. |
graph TD
M["MAST v1.0"]
M --> A["Agent workspace"]
M --> B["Model layer"]
M --> C["Memory layer"]
M --> D["Automation layer"]
M --> E["Security layer"]
M --> F["Voice and vision layer"]
A --> A1["OpenCode, Cursor, VS Code, Windsurf"]
A --> A2["MCP server hub"]
B --> B1["Cloud provider routes"]
B --> B2["Local Ollama fallback"]
C --> C1["ChromaDB semantic memory"]
C --> C2["SQLite task ledger"]
D --> D1["Playwright browser control"]
D --> D2["Scheduler and notifications"]
E --> E1["Authorized OSINT tooling"]
E --> E2["Nmap, Shodan, Nuclei workflows"]
F --> F1["Whisper speech input"]
F --> F2["OCR and screen understanding"]
|
Unified AI stack merging M4STCLAW, OpenWork, and EIGENT into a local-first operator: 21 MCP servers, 11 provider routes, task chains, semantic cache, memory, scheduler, and local fallback.
|
AI-assisted OSINT and pentest workflow with Nmap, Shodan, Nuclei, target-scope guardrails, and CEH-style methodology for defensive research and authorized testing.
|
|
Portable MCP workspace stack for AI IDEs. Focused on safe file work, browser automation, research, memory, skills, notifications, and tool reconstruction through config.
|
Local outreach pipeline for scraping public directories, filtering prospects, scoring by ICP rules, and drafting personalized emails through LLM-assisted workflows.
|
|
Also comfortable with: REST APIs, JSON-RPC, CLI tooling, Windows automation, Linux workflows, config-driven systems, and repo packaging. |
Model routes: Groq, Cerebras, Gemini, DeepSeek, Kimi K2, Qwen/Qwen-Coder, Together AI, Sarvam-M, Nemotron, and local Ollama fallback. |
|
Workspace: Windows AtlasOS, Kali Linux dual boot, OpenCode, Cursor, VS Code, Windsurf, browser automation, and reproducible setup scripts. |
Focus areas: authorized recon, target-scope validation, browser fingerprint testing, web automation, vulnerability template workflows, and defensive research tooling. |
Open to: AI automation work, MCP integrations, OSINT tooling, agent workflows, and serious open-source collaboration.
Best fit: practical systems that need to run locally, cheaply, and reliably.
Preferred contact: email or LinkedIn DM.