Security Engineer | OT/ICS & Cloud Detection Engineering
I build end-to-end security architectures for critical infrastructure. My focus is on bridging the gap between legacy industrial protocols (Modbus, S7comm) and modern cloud-native detection pipelines.
LinkedIn · Email · Based in Spain (Open to Remote Europe/USA)
I don't just build security tools; I engineer the data pipelines required to ingest, structure, and analyze high-volume industrial telemetry for advanced heuristics and ML modeling.
graph TD
subgraph "On-Prem / Edge (OT-Security-Lab)"
PLC[PLCs / HMI] -->|Modbus/TCP| GW[OT Gateway / NDR]
GW -->|Suricata/Zeek| ML[Malcolm NDR Pipeline]
end
subgraph "Cloud Telemetry Lake (AWS/LocalStack)"
ML -->|Fluent Bit| S3[S3 Raw Storage]
S3 -->|SQS/Lambda| LP[Log Parser Toolkit]
LP -->|Detections| DDB[(DynamoDB)]
DDB -->|Streams| IR[Automated NIST Reports]
end
style GW fill:#f96,stroke:#333
style LP fill:#ff9,stroke:#333
style IR fill:#dfd,stroke:#333
The Problem: SOC analysts in OT environments drown in high-noise alerts. Generating NIST-aligned incident reports and Suricata rules manually is slow, inconsistent, and doesn't scale. The Solution: An agentic AI pipeline that detects anomalies via Isolation Forest, enriches them with RAG-augmented OT knowledge (IEC 62443, asset inventories, past incidents), and produces NIST SP 800-61 reports with custom Suricata rules — all without human intervention.
- Engineering Challenge: Built a deterministic classification layer that routes alerts to the correct analysis path before LLM invocation, eliminating token waste. Made the agent LLM-agnostic — swap between GPT-4o-mini and local Ollama models by changing two env vars, no code changes needed.
- Stack: Python, LangChain/LangGraph, ChromaDB, FastAPI, scikit-learn, OpenAI/OpenRouter, Pytest. 25 deterministic tests pass in CI without API keys.
The Problem: Ingesting OT telemetry into AWS is often rigid and expensive. The Solution: A serverless, event-driven pipeline that ingests, parses, and archives OT security events in real-time.
- Engineering Challenge: Solved LocalStack Community constraints by implementing a Fat-Zip dependency injection at cold-start and a dynamic gzip detection layer for Fluent Bit payloads.
- Stack: Terraform, AWS Lambda, DynamoDB, S3, Snappy/Parquet, Fluent Bit.
The Problem: Commercial NDR (Nozomi/Claroty) is cost-prohibitive for many facilities. The Solution: A production-grade NDR pipeline using CISA Malcolm, Arkime, and Suricata, enriched by a custom Python SOAR layer.
- Impact: Implemented automated DPI profiling of Modbus function codes to identify unauthorized register manipulation before it hits the SIEM.
- Stack: CISA Malcolm, Arkime, Suricata, Python (Scapy/Tshark).
The Problem: You can't test attacks on live water treatment plants. The Solution: A 5-zone Docker-based simulation of a water filtration facility mapped to the Purdue Model and IEC 62443.
- Engineering Judgment: Isolated the Historian in Level 3 to enforce unidirectional data flow, fulfilling IEC 62443 requirements for zone-to-zone restricted access.
- Stack: OpenPLC, Scada-LTS, Iptables (Zone Firewall), InfluxDB, Grafana.
The Problem: SIEM ingestion is only as good as its parser. The Solution: A memory-efficient, stateful parsing engine for unstructured logs.
- Technical Nuance: Uses the Generator pattern to process multi-gigabyte logs with near-zero RAM overhead. Features a stateful middleware for correlating SSH brute force and web scanning across time windows.
- Availability is Paramount: In OT, a False Positive that triggers a block can be more dangerous than the attack itself. I focus on high-fidelity, physics-aware detection.
- Threat-Informed Defense: Every detection rule I write is mapped to MITRE ATT&CK for ICS (T0831, T0846, T0886) to ensure coverage of actual adversary TTPs.
- Evidence over Opinions: I value raw PCAPs, verified attack logs, and NIST-aligned incident reports over "box-ticking" compliance.
- Data Quality is Paramount: I focus heavily on the data engineering lifecycle. A predictive model in OT is useless without low-latency, highly structured telemetry.
- Physics-Aware Modeling: False positives in ICS cost downtime. I emphasize high-fidelity feature engineering (e.g., mapping TTPs to MITRE ATT&CK for ICS) to ensure models understand actual industrial context, not just statistical noise.
- Research & Application: Continuously researching the intersection of Deep Learning and Cybersecurity, including time-series anomaly detection and integrating LLMs (RAG) for automated incident response contextualization.
- Aligning with Industry Standards: Actively hardening expertise via GICSP (Global Industrial Cyber Security Professional), BTL1, and Security+.
- Focus: Advancing my knowledge in Cloud-Native SIEM (Sentinel/Chronicle) and ICS Adversary Emulation (TRITON/Industroyer).
While my full-time focus is defending critical OT architectures, I spend my evenings and weekends immersed in the applied AI/ML community.
I am highly active in the broader engineering space and am always open to connecting regarding joint research initiatives, open-source collaborations, and technical advisory on data engineering and predictive modeling challenges. Whether it's architecting a robust data pipeline or exploring models for anomaly detection, feel free to reach out!