Hugo Perrier, PhD

Regex Artist | Machine Learning Engineer | NLP Practitioner

Niort, Nouvelle-Aquitaine, France

About

Core contributor and maintainer of the Open-Source package Melusine for automatic email processing. Data enthusiast with a background in physics and computational fluid dynamics (CFD).

Experience

  • Data Scientist / Expert NLP at MAIF
    Apr 2023 - Present · 3 yrs 4 mos

  • Enseignant Vacataire at BUT - Science des Données - Niort
    Jan 2023 - Present · 3 yrs 7 mos

    Give a Clustering (Unsupervised Machine Learning) course to 2nd year Data Science BUT students. - Prepare lecture slides and interactive material - Prepare practical work activities using Python and Google colab

  • Instructor - Machine Learning Engineer Bootcamp at Yotta Academy
    Feb 2020 - Jul 2023 · 3 yrs 6 mos

    La Yotta Academy est la première formation en France dédiée au métier de Machine Learning Engineer. Instructeur dans le Track Natural Language Processing (NLP)

  • ML Engineer at Quantmetry, Building AI with pioneers
    Sep 2018 - Mar 2023 · 4 yrs 7 mos

    Core contributor to the Open-Source NLP package Melusine for automatic processing of french emails. - Development of NLP functionalities such as Semantic Analysis in the Melusine package - Maintenance of the Open-Source package (merge requests, documentation, tutorials, continuous integration, etc) Productionized implementation of Melusine in the Information System of a major insurance company (10k+ emails processed / day) - Emails routing (based on NLP analysis of the email content) - Emails prioritization / deprioritization (Ex: emergency detection) - Speed-up email processing with auto-responses (Ex: reply to document requests) - Code optimization to speed-up API response time (ensure response time < 100 ms) Failure diagnostic for industrial equipments in the aeronautics industry - Development of a predictive model based on a Bayesian statistics methodology - Failure probabilities calculation using numerical simulation and Markov Chain Monte Carlo - Integration of the solution developed into the domain experts’ industrial process - Development of an ergonomic graphical user interface tailored to the needs of equipment operators Failure diagnostic for industrial equipments in the defense industry - Characterization of failure patterns - Development of a model to distinguish real failures from test equipment failures - Focus on model interpretability to communicate and discuss model findings to domain experts

  • Researcher in Fluid Dynamics Code Development at Imperial College London
    Oct 2014 - Oct 2018 · 4 yrs 1 mo

    Develop numerical methods to model complex Nuclear flows using Computational Fluid Dynamics (CFD) In the event of a nuclear accident, the reactor core can melt down and form a high temperature "lava like" liquid called corium. If the corium is not cooled down soon enough, it could melt through the reactor building concrete basemat and harm populations and the environment. Numerical simulations can help design safer reactors and effective accident management measures. • Leading an open-ended research project on the development of numerical methods to simulate corium flows. • Collaborating with a 6 people team to develop a parallel scientific code (C, MPI). • Presenting research results and implications to panels of experts and at conferences.