Noé Beserman

Independent technical advisory for companies deploying production AI systems. CURRENT CLIENTS & PROJECTS (2023-present) - Neuroute (London): Healthcare AI & voice synthesis — clinical trial optimization, medical document NLP, production deployment (<100ms latency). See detailed project below. - Diot-Siaci (insurance): AI-powered document processing and risk analysis automation — NLP extraction, classification, production deployment. - SCSE: Automated reporting pipeline with ML-powered data consolidation, PowerPoint generation, Power BI integration. EXPERTISE - LLM deployment: RAG architectures, fine-tuning strategies (GPT-4, Claude, Llama, Mistral), prompt engineering - MLOps: Kubernetes, Docker, CI/CD, monitoring, model versioning - Production ML: Scalability, performance optimization, <100ms latency systems - Technical leadership: Architecture decisions, team guidance, C-level communication - Strategic consulting: Technical audits, ML architecture, POC feasibility studies, deployment strategy - Training workshops: GenAI adoption, LLMOps best practices STACK Python • PyTorch • Langchain • LlamaIndex • HuggingFace Transformers • FastAPI • Kubernetes • Docker • PostgreSQL • Redis • AWS • Azure

Logo de Neuroute Senior AI Engineer (Freelance) Senior AI Engineer (Freelance) Neuroute · FreelanceNeuroute · Freelance avr. 2024 - nov. 2024 · 8 moisDu avr. 2024 au nov. 2024 · 8 mois AI engineering for a healthtech company specialized in clinical trial optimization. DEEP LEARNING VOICE SYNTHESIS - Architected and deployed production-grade voice synthesis system for healthcare applications - Technologies: WaveNet, Tacotron 2, real-time inference pipeline - Performance: <100ms latency for clinical use cases - Stack: PyTorch, TensorFlow, Docker, FastAPI GRAPH-BASED CLINICAL OPTIMIZATION - Designed multi-layer graph analysis framework (Neo4j) for patient recruitment and site selection in clinical trials - Significantly reduced trial planning cycles through network optimization algorithms - Applied complex systems methodologies to healthcare logistics MEDICAL NLP PIPELINE - Built ML pipeline for medical document analysis: entity extraction from clinical notes, automated patient data structuring, adverse event detection from unstructured medical data - Technologies: HuggingFace Transformers, spaCy, custom NER models STACK PyTorch • TensorFlow • Neo4j • FastAPI • Docker • HuggingFace Transformers • Medical NLP • Graph Analytics

Explored startup opportunities leveraging AI, statistical mechanics, and complex system theories. Developed and validated innovative business models, pitching to investors and mentors.

ENS Ulm (sciences, promotion 2017). Completed rigorous academic training in physics, specializing in statistical mechanics and econophysics. Conducted interdisciplinary research bridging physics, economics, and computational modeling.

Generalized Schelling's segregation model through spin glass physics. Applied statistical mechanics methodologies to economic and public policy phenomena.

Developed strategic scenarios for carbon transition and sustainability planning. Modeled emissions trajectories (2030-2050) impacting global automotive markets. Liaised with OECD, International Energy Agency, and International Transport Forum.

Built predictive models linking commodities and corporate data. Constructed graphs and metrics for synthetic data over Wikipedia automated extraction. Developed mathematical models forecasting stock market dynamics.

Conducted research in econophysics, creating econometric factors and mathematical models analyzing the U.S. economy. Developed inference models for systematic sector decomposition of companies.

Experimental and theoretical study of non-Newtonian fluid dynamics. Developed models for fluid interaction with textured surfaces.