Luca M.-Senior AI Engineer · Production LLM Systems & Applied Research

UK-based applied AI lab for autonomous security analysis. Remote contract engagement.

• Post-training of open-weight LLMs for the company's autonomous security-analysis product (AI integration, deployment on customer infra): domain-specific abliteration and fine-tuning to raise performance on specialised cybersecurity tasks.
• Co-built RedLineBench (open, MIT): 153 grounded tasks scoring refusal and capability separately, weighted LLM-as-judge rubric, reference run across frontier and open models with two independent judges. Agentic benchmark in preparation.

Repository: [link] Write-up: [link]

Co-founded an AI e-commerce optimisation platform; raised €1.1M pre-seed.

Built an agentic pipeline that rewrites e-commerce product listings at scale, plus cross-encoder product classification at 95%+ accuracy on an open-set taxonomy over 100K+ products.

B2B engagements alongside the roles above; long-running and project-based clients.

Aurelio AI — since Oct 2023, long-running engagement

• Core contributor to semantic-router, Aurelio's open-source decision layer for LLM apps (deterministic routing, cost/latency control) — 3,200+ stars, used in production by third parties.
• Built retrieval, routing and evaluation components for enterprise RAG systems; shipped NLP solutions end to end.

PolarFrequency, WritePal, Muze, XRIGLOBAL — project-based engagements

• Design, ship and operate production LLM systems (RAG pipelines, agents, multimodal workflows) serving 600K+ MAU across clients, owning architecture, evaluation, deployment and observability end to end.
• Typical engagement: take an LLM feature from prototype to production — retrieval and routing design, eval sets to catch regressions, cost/latency budgeting, monitoring — and advise founders on the AI roadmap.

Applied AI research for robotics: led the CV division, 3x faster YOLOv5/v8 inference on NVIDIA Jetson, SLAM and ROS2 integration for autonomous robots.