Research Scientist – Foundation Models and Multimodal AI

<h1>About us</h1><p style="min-height:1.5em">Owkin is an agentic AI company pioneering Biological Artificial Superintelligence to solve problems in biology where human researchers alone have failed.<br />Owkin builds K Pro - an AI scientist for pharmaceutical research and strategic decision-making. K Pro orchestrates a suite of AI skills and tools to decode complex biology, accelerate research, and dramatically increase productivity.<br />K Pro is built on Owkin’s unrivalled multimodal patient data network, state-of-the-art AI for biology and a decade of experience working with pharmaceutical partners.</p><p style="min-height:1.5em"><em><strong>Position is based in our Paris office or remotely in France, UK, Germany.</strong></em></p><p style="min-height:1.5em"><em><strong>Please submit your CV in English</strong></em></p><p style="min-height:1.5em"></p><h1>About the role:</h1><p style="min-height:1.5em"> We are seeking a driven <strong>Research Scientist</strong> with hands-on expertise in Foundation Models (FMs) and Multimodal AI to help drive the next phase of our machine learning platform.</p><p style="min-height:1.5em">In this role, you will bridge cutting-edge AI research and practical application. Rather than training massive base models from scratch, your focus will be on fine-tuning, adapting, and aligning open-weights foundation models across diverse data modalities (e.g. structured tabular data, single-cell scRNA-seq, image representations, and natural language) to build multimodal systems.</p><p style="min-height:1.5em"> </p><h1>In particular, you will be responsible for:</h1><ul style="min-height:1.5em"><li><p style="min-height:1.5em"><strong>Model adaptation and optimization:</strong> Adapt, fine-tune, and evaluate state-of-the-art open-source foundation models (e.g., domain-specific transformers, vision-language models) using fine-tuning and transfer learning strategies.</p></li><li><p style="min-height:1.5em"><strong>Multimodal integration:</strong> Design and implement cross-modal alignment architectures (e.g., contrastive learning, projection heads, unified embedding spaces) to fuse structured, sequential, and text/image data layers.</p></li><li><p style="min-height:1.5em"><strong>Pipeline and evaluation engineering:</strong> Build rigorous, domain-tailored benchmarking frameworks and evaluation pipelines to validate model reliability, generalization, and downstream task performance.</p></li><li><p style="min-height:1.5em"><strong>Applied R&amp;D collaboration:</strong> Work closely with cross-functional teams (data engineers, domain specialists, and software engineers) to integrate adapted models into scalable production and inference workflows.</p></li><li><p style="min-height:1.5em"><strong>Low-hanging fruit innovation:</strong> Identify high-impact, incremental improvements in data pre-processing, embedding representation, and fine-tuning to maximize model performance without unnecessary computational overhead. </p></li></ul><p style="min-height:1.5em"> </p><h1>About you</h1><p style="min-height:1.5em">Required qualifications / experience:</p><ul style="min-height:1.5em"><li><p style="min-height:1.5em"><strong>Education:</strong> Ph.D. or Master’s degree in Computer Science, Machine Learning, Mathematics or a related quantitative field.</p></li><li><p style="min-height:1.5em"><strong>Experience:</strong> 2-4 years of post-degree experience applying deep learning and foundation models to complex datasets.</p></li><li><p style="min-height:1.5em"><strong>Foundation models:</strong> Demonstrated experience adapting and fine-tuning foundation models.</p></li><li><p style="min-height:1.5em"><strong>Multimodal learning:</strong> Proven experience with cross-modal representation learning, joint embedding alignment (e.g., CLIP-style architectures), or vision-language models.</p></li><li><p style="min-height:1.5em"><strong>Technical stack:</strong></p><ul style="min-height:1.5em"><li><p style="min-height:1.5em">Advanced proficiency in Python and modern deep learning frameworks (<strong>PyTorch</strong>, Hugging Face ecosystem).</p></li><li><p style="min-height:1.5em">Hands-on experience with fine-tuning techniques.</p></li><li><p style="min-height:1.5em">Familiarity with data engineering tools for handling large-scale or multi-modal datasets (Parquet, AnnData, HDF5).</p></li></ul></li></ul><p style="min-height:1.5em"> Preferred qualifications/bonus:</p><ul style="min-height:1.5em"><li><p style="min-height:1.5em">Experience working with biological or omics data (e.g., scRNA-seq, spatial transcriptomics, clinical and/or histology data).</p></li><li><p style="min-height:1.5em">Experience deploying or scaling models using containerized environments (Docker, Kubernetes) and cloud infrastructure (AWS/GCP).</p></li><li><p style="min-height:1.5em">Track record of open-source contributions or peer-reviewed publications at top ML conferences (NeurIPS, ICML, ICLR, CVPR) or domain-specific journals.</p></li