Youssef Lamin

I design and build AI systems that work — not just demos. My focus is on multi-agent orchestration, LLM evaluation, and constraint-driven model optimization.

What I offer:
• Multi-agent workflow design & orchestration (AWS Strands Agents SDK, tool-calling, state management)
• LLM evaluation frameworks — finding blind spots, hallucinations, and failure modes (LLM-as-a-Judge, A/B testing, benchmarking)
• Model optimization for edge & constrained environments (0.13 MB CNN ranked top 2.3% globally in Arm Create)
• AI research & technical writing — I publish methods, not just results

What makes me different:
• I think in systems, not sentences — I engineer behavior, measure it, and document its failure modes.
• I publish unflattering results. If a model doesn't work, I say so and explain why.
• I bridge creative intent and machine execution — from narrative design to agent architecture.

Track record:
• G.T.B v2 — multi-agent orchestrator built on AWS Strands Agents SDK (AWS Hackathon)
• LLM-snake — comparative evaluation of 5 LLMs on hidden pattern detection
• Opt-MobileNet — top 2.3% (57/2476) in Arm Create AI Optimization Challenge
• 3 research publications indexed on Zenodo (CERN) with permanent DOIs

I'm available for freelance projects, consulting, and research collaborations. Remote worldwide, GMT timezone.