At LightOn, we are building Paradigm, a sovereign Private Generative AI platform designed for Fortune 500 enterprises who prioritize data privacy and intellectual property protection. Our platform enables secure interactions with corporate documentation while safeguarding organizational know-how. Together, we’re reshaping how industry leaders leverage Gen AI while maintaining complete control of their critical information assets.
Our Machine Learning (ML) teams work end-to-end on the training, inference, and evaluation of transformer models and retrieval systems that power Paradigm – our enterprise Generative AI Copilot. We also undertake custom training projects for clients, public initiatives, and our own research endeavors such as ModernBERT. We are committed to advancing the field through open-source contributions and publishing research when appropriate, sharing our innovations with the broader AI community.

As a CIFRE PhD candidate, you will drive cutting-edge research at the intersection of academia and industry, advancing LightOn’s capabilities in generative AI and retrieval systems. You will:

• Define and lead research initiatives in collaboration with ML teams and academic supervisors, focusing on areas such as:

  • Novel data creation and curation methodologies

  • Sample-efficient training strategies for model customization

  • Comprehensive evaluation frameworks for generative models and retrieval systems

  • Advanced retrieval architectures and knowledge management

  • Model compression and optimization for production deployment
    Translate research into impact by publishing in top-tier conferences and journals, contributing to open-source projects, and presenting findings to both technical and business stakeholders

  • Participate in collaborative research projects with academic institutions and public initiatives

  • Bridge the gap between theory and practice by maintaining awareness of latest ML developments while ensuring research aligns with LightOn’s strategic goals and real-world applications

    Your Qualifications

    Essential

  • Master’s degree in CS, Applied Math, Physics, or equivalent with strong theoretical foundations

  • Understanding of transformer architectures, attention mechanisms, and modern NLP techniques

  • Knowledge in Python and PyTorch

    Appreciated

  • Demonstrated research capability through publications, preprints, or substantial open-source contributions

  • Prior work on LLM evaluation, benchmarking, or metric design

  • Experience with information retrieval

  • Track record of building end-to-end ML systems that made it to production

    If you don’t meet all the criteria but are passionate about advancing AI research through rigorous scientific work, we encourage you to apply. We value potential and dedication as much as experience.

1st stage: initial alignment interview

2nd stage: technical interview

3th stage: a call with top management