Florian Laurent

Florian Laurent is a Machine Learning engineer and entrepreneur. He works on Generative AI and Reinforcement Learning. He is based in Switzerland (Vaud).

NyxAI (2022-current) Co-founder of a Generative AI startup providing state of the art tooling for image generation. Our product dreamlook.ai allows users to finetune Stable Diffusion text-to-image models in minutes.

Cedille.ai (2021-2022) Trained and deployed the largest language model in French (6b), released under an open-source license (MIT), with a free and public playground.

ML engineer at Coteries (2020-2022) I worked on ML projects, both for clients and internally, with a focus on NLP/NLG. Worked on the development, training and deployment of LLMs (10b+ parameters) leveraging large GPU and TPU clusters.

AIcrowd (2019-2021) Lead researcher of the Flatland project in 2020, which investigates the use of ML (specifically Deep Reinforcement Learning) for railway planning, in partnership with the national railway companies from Switzerland, Germany and France (SBB, Deutsche Bahn, SNCF).

Co-organizer of the associated Flatland 2020 NeurIPS Challenge. The challenge received >2k user submissions with results presented and published at NeurIPS 2020.

Workshop organizer: I presented highly-rated workshops about RL and NLP:

  • “Meet your Artificial Self: Generate text that sounds like you” at AMLD2020
  • “Climbing the Ladder: Reinforcement Learning in a Competitive Setting” at AMLD2020
  • “Flatland: Multi-Agent Reinforcement Learning on Trains” at TheWebConference 2021

Full Stack/Machine Learning Freelance (2017-2019) I worked as a full-stack web developer, using both the enterprise stack (Oracle/Hibernate/Spring) and the startup stack (React/Express/Firebase). I also have experience with Android native development, with both Java and Kotlin.

Selected mandates over this period:

  • Development of an Android application for the TV box of a national operator, deployed to >1M users (3 months full-time mandate, 2017)
  • Setup a 128 GPUs Kubernetes cluster on AWS for the final evaluations of the NeurIPS 2018 Adversarial Vision Challenge competition (4 months 60% mandate, 2018)
  • Conception of a Deep Reinforcement Learning course for a private EPFL-based company (12 months 60% mandate, 2019)

Full Stack Software Engineer at Swissquote Bank (2015-2016) I worked on the development and maintenance of the “information” services of the bank including financial charting tools, search engines, alerts for prices and news - both backend (Java, J2EE, Spring, Stripes) and frontend (HTML5, CSS, JS).

CERN Software Engineering Intern (2014) I built a web-based data visualization system for the ATLAS experiment. My tool was designed to ingest >1M events per second per node and summarize them in dashboards to be used by experts. It was my EPFL Master thesis and it received maximal grade.

My solution consisted of a web-based Angular frontend and Java backend. The frontend consisted of a Grafana fork, using a new graph rendering engine, optimized communication protocol and a revamped user interface developed based on user feedbacks. The backend took care of ingesting up to 1M events per second, storing them in sharded Redis nodes, and pre-generating downsampled views for faster access from the frontend.

Last updated June 16 2023