AI ENGINEER · AGENTIC AI & LLMOPS · DATA SCIENTIST
AI Engineer specializing in production-grade LLM systems: agentic AI, evaluation and observability, LLMOps. Former industrial engineer, I apply the discipline of aerospace production lines to LLM systems: measured, observable architectures, built for production.
Ask an AI who I am:
My career began in aerospace at Safran, in methods and then industrialization. Convinced of what data could do for industry, I moved into data science, then AI engineering, ultimately operating as founding engineer at a pre-seed AI startup. Today I design production-grade LLM systems: agents, RAG, evaluation and observability. This dual background, industry and AI, gives me systems thinking and a demand for reliability on every project.
My experienceAgents, RAG, evaluation: from raw data to production LLM systems.
Production-grade LLM systems: multi-agent orchestration with LangGraph and MCP, RAG and document-intelligence pipelines, evaluation and observability (Ragas, LangSmith), deployment on AWS Bedrock or vLLM.
Predictive models and deep learning, from feature engineering to deployment: Python, Scikit-Learn, TensorFlow, NLP and time series in service of actionable decisions.
End-to-end industrialization: data pipelines (Spark, Kafka), CI/CD, Docker and MLflow, AWS, GCP and Azure infrastructures built for reliability and scalability.
Built as products, proven like industrial systems: agentic and generative AI, from R&D to production.
A LangGraph multi-agent architecture: Plan-and-Execute, ReAct sub-agents, human-in-the-loop and measured Bedrock prompt caching, 51.7% lower multi-turn cost.
Multimodal RAG pipeline on Databricks and a Ragas + LangSmith evaluation framework: the in-house pipeline beat the commercial baseline, P@k 0.82 vs 0.68.
AI UGC video production engine: LLM-orchestrated generation, vision quality gates, three-layer cost observability. Around $3.5 per publish-ready video.
An AI agent over real-time industrial telemetry: production KPIs answered in natural language, quality-drift diagnosis, Lean Six Sigma meets generative AI.
Marius was a founding engineer at Arkim, where he designed and deployed core parts of our AI platform. Including our multimodal RAG pipeline and agentic workflows built on LangGraph. Working across time zones with minimal oversight, he consistently took ambiguous problems and came back with shipped, production-ready solutions. I'd bring him in again without hesitation.See on LinkedIn
Working alongside Marius at Arkim has been at absolute pleasure and I have no doubt that he will continue to excel in any technical role that he chooses to pursue. Like myself, he's a guy who came from heavily regulated industry (aerospace, in his case) and developed a passion for building reasoning systems with AI tools that solve real-world problems. As CTO, I worked very closely with Marius developing agent orchestration tools and context management systems where I witnessed the following first-hand: 1. he is brilliant 2. he produced the best architecture diagrams and test reports I've seen in my career 3. he's very quick to learn, innovate, and adapt I sincerely hope to work together again in future projects and would highly recommend Marius to any team in need of AI expertise.See on LinkedIn
Let's talk about your needs: LLM system design, data science or model industrialization.
EN
FR