AI systems builder · AI Security & offensive security (in training) · Law enforcement professional · Ex-PLC programmer (Omron, international deployments)
Built 18+ production AI systems independently in 6 months — FinTech, GovTech, industrial RAG, local AI infrastructure, and RPA automation. Now specializing in AI Security / LLM red teaming and ethical hacking.
I come from industrial automation (PLC programming, international deployments in Pakistan, Madrid, Barcelona) and 18 years in law enforcement. Since late 2025 I've been applying that systems-thinking mindset to AI — building real, deployed tools that solve real operational problems — and I'm now moving into offensive security, where my AI background and systems experience meet.
No tutorials. No toy projects. Everything here runs in production or was built for a specific operational need.
| Project | What it does | Status |
|---|---|---|
| METATRON | AI-powered pentesting assistant (CLI). Automates the engagement cycle: recon → vulnerability-pattern recognition → advisory attack-tree. A local LLM writes the report while the code owns the ground truth (version-filtered searchsploit cross-referencing) with anti-hallucination guardrails. Verified live against 10+ rooted lab machines (web exploitation, FTP, privilege escalation chains) | Production-tested |
| Pantocrator | AI exploitation-advisory engine. Ingests recon data, generates a prioritized attack plan from a verified 64-technique playbook (33 initial access + 31 privilege escalation) using a local ablated LLM. Hard safety boundary by design: the system advises, a human executes — no autonomous exploitation. SQL-backed learning loop ranks techniques by real success rate | In development |
Focus: AI Security / LLM red teaming (OWASP Top 10 for LLM) · web & network pentesting
Learning path: eJPT → PNPT → OSCP
Stack: Kali Linux · nmap / rustscan · Metasploit · searchsploit · web exploitation (SQLi/LFI/upload) · privilege escalation · Docker labs
| Repo | What it does | Status |
|---|---|---|
tingbot_binance |
Algorithmic trading bot on Binance (BTCUSDC, M1). 6-module pipeline: signal detector → thermometer → weather model → decision engine → watchdog → capital manager. 24/7 autonomous operation with Telegram alerts | Production |
tickbot |
Tick-level signal processor and order manager | Production |
trading-agent-v1 |
First-generation trading agent with real-time chart generation and automated signal visualization | Production |
Stack: Python · Binance API · systemd · Hetzner VPS · Telegram Bot API
Built for real public-administration operations. Source and data are private (sensitive operational systems) — happy to walk through the architecture in an interview.
| Project | What it does | Status |
|---|---|---|
bitacola |
Operations-management platform for a municipal police force. Self-hosted (Debian · Nginx · PM2 · SSL · private mesh VPN) | Production · Private |
sherlock |
AI document-generation engine — the officer dictates and an LLM produces legally-formatted official reports in real time (FastAPI) | Production · Private |
penal |
Legal-code lookup and behavioural pattern analysis using embeddings + cosine similarity | Production · Private |
central-partes |
Unified command panel: photo ingestion, AI cataloging, auto-formatting to the official report standard | Production · Private |
iactes |
Digital creation and management of municipal infringement records, with integrated reference management | Production · Private |
photo-evidence-manager |
Desktop app that cross-references field photos with case CSVs; automatic file indexing (Watchdog) | Production · Private |
legacy-migration |
Migration of a legacy Access/VBA municipal management system (~40 tables) to Python + SQLite | Legacy / Migration · Private |
OPOS |
Daily scraper of public municipal notice boards (taulell d'edictes) — automated detection of civil-service exam announcements with instant alerts | Production |
firmadoc-monitor |
Playwright automation of a municipal BPM platform: daily monitoring of records, Excel extraction, Telegram alerts | Development · Private |
Stack: Python · Node.js · FastAPI · Express.js · SQLite · docxtemplater · python-docx · Playwright · Tkinter · Watchdog · Anthropic API · Nginx · PM2 · Telegram Bot API
| Repo | What it does | Status |
|---|---|---|
bafatec-rag |
Production RAG system over ~5,000 industrial engineering project files — vector search, receipt OCR/cataloging, and automated technical report generation | Production |
whatsapp-to-project |
Processes exported WhatsApp conversations (audio + text) to auto-extract and structure the data for a full electrical-engineering project. Whisper STT + domain-specific LLM extraction pipeline. Validated with local Llama and Nvidia API | Local / demo available |
Stack: Python · Embeddings · Vector DB · OCR · Whisper/STT · Anthropic API · Document generation
| Repo | What it does | Status |
|---|---|---|
jarvis-cognitive |
Local cognitive assistant on Ollama (Llama 3.2 / Qwen 2.5). On-premise, zero data egress. GPU node available (RTX-class) for scaling | Active |
Stack: Ollama · Llama 3.2 · Qwen 2.5 · Python · Linux Debian · Local-first architecture
| Repo | What it does | Status |
|---|---|---|
Alpha50-core |
Personal fitness and weight tracking agent — MVP complete, on standby | Standby |
Self-hosted production infrastructure (Linux Debian · Nginx · systemd · PM2 · SSL · private mesh VPN) plus a GPU inference node (RTX-class) for local LLMs.
Cloud platforms: AWS (EC2 · 24/7 trading bots) · Google Cloud · Hetzner VPS
LLM Integration ████████████ Production (Anthropic API, Ollama)
RAG / Vector Search ████████████ Production (5,000+ file corpus)
Agent Orchestration ██████████░░ Multi-agent systems
Web Automation / RPA ████████████ Playwright, Watchdog, scraping, BPM
Server Deployment ████████████ AWS · Google Cloud · Hetzner · self-hosted Debian
API Development ████████████ FastAPI, Express.js, REST, webhooks
Document Generation ██████████░░ Word/docxtemplater, AI-personalized records
Audio / STT Pipeline ████████░░░░ Whisper, conversation parsing, data extraction
Trading Systems ████████░░░░ Binance API, 6-module signal pipeline
Local LLM / GPU ████████░░░░ Ollama, RTX-class node (expanding)
AI Security / LLM RT ██████░░░░░░ Building (OWASP Top 10 for LLM, MITRE ATLAS, Pantocrator safety design)
Offensive Security ████████░░░░ 10 machines rooted — SQLi, LFI+log poisoning, SSTI, JWT, 9 GTFOBins vectors, pivoting (eJPT prep)
Industrial Automation ████████████ PLC Omron (Ladder), SCADA, Syswin
Database Design ████████████ SQLite, relational design (40+ table systems)
Languages: Python · Node.js · JavaScript · SQL · VBA/Access · Visual Basic · Ladder (PLC Omron) Security: pentesting (web & network) · nmap/rustscan · Metasploit · searchsploit · privesc · LLM security Cloud: AWS (EC2) · Google Cloud · Hetzner VPS · private mesh VPN Infrastructure: Linux Debian · Nginx · systemd · cron · SSH · PM2 · Docker AI/ML: Anthropic API · Nvidia API · OpenAI API · Ollama · Whisper · Prompt engineering · RAG · Agents
- 🚓 Active law enforcement officer — Policia Local, Catalonia (18 years, including team lead)
- 🏭 Former PLC programmer — Omron automation, international deployments (Pakistan, Madrid, Barcelona)
- 📐 Physics degree (UNED) — strong mathematical and analytical foundation
- 🥋 1st Dan black belt, Taekwondo
- 🌍 Languages: Catalan (C1 certified) · Spanish (native) · English (technical/intermediate)
Roles where domain expertise meets AI engineering and security:
- AI Security Engineer / LLM red teaming — securing and testing AI systems
- Penetration Tester (junior) — offensive security, ethical hacking
- AI Engineer — building and deploying LLM-based systems
- Automation / RPA Engineer — intelligent process automation
- GovTech / LegalTech — AI for public administration or legal workflows
Open to remote. Based in Catalonia, Spain. Certifications in progress: AWS Certified AI Practitioner (exam scheduled) 📫 Reach me via GitHub or LinkedIn.