AI Red Teamer & Automation Architect

Miguel
Automate

Transforming operations through secure AI integration, architecting resilient cloud infrastructure, and building bespoke automation solutions that scale.

View Projects Get in Touch

About Me

Red teaming the future of AI.

As an AI Red Teamer and Automation Architect, I work at the intersection of generative AI and offensive security. I do not just build agents. I find breaking points in logic, leaks in data flows, and vulnerabilities in orchestration layers so businesses can scale with LLMs without sacrificing security or operational integrity.

I start with a deep dive into existing business processes to identify manual, repetitive, or error prone tasks, then design bespoke automation solutions primarily using Julia. My philosophy is about crafting tailored software that integrates seamlessly with current systems.

I specialize in AI red teaming for audit reports, including jailbreaks, prompt injection, data poisoning, and insecure output handling in complex RAG pipelines. The outcome is clear, actionable findings that harden AI systems.

Adversarial Testing for LLM Systems

Prompt Injection & RAG Risk Analysis

Audit Mitigations & Retests

Tech Stack

Tools and platforms I use to build secure, scalable automation systems.

Offensive AI & Red Teaming

PyRIT Garak Giskard ART (Adversarial Robustness Toolbox) Prompt Injection (Direct/Indirect) Jailbreaking Model Inversion Data Poisoning

LLM Evaluation & Auditing

PromptFoo RAGAS DeepEval TruLens

Orchestration & Agents

LangChain CrewAI AutoGen LlamaIndex Agency Swarm n8n Composio SuperAGI LangGraph Semantic Kernel Langflow MCP

Network & Infrastructure Security

TCP/IP Wireshark Nmap Metasploit Burp Suite pfSense AWS Azure GCP Docker Kubernetes Terraform Ansible Bash/Batch

Featured Services

Independent AI red teaming for MCP and multi-agent systems. Safety-first, black-box audits without code access or prompt sharing.

AI Pen Testing

Agentic Security Audits (AI Red Teaming)

I test your agents the way an external attacker would—focusing on where context changes intent, where handoffs break, and where tools can be abused. You receive an evidence-first report with reproducible PoCs (TRACE → BREACH → IMPACT → PROOF), a trust-boundary map, OWASP LLM Top 10 mapping, and an executive summary your team can act on fast.

Learn More