AI Research Scientist (Multimodal post-training)

Design and execute research on multimodal model training — with a primary focus on vision-language models and, increasingly, speech-language models — including fine-tuning, alignment, and post-training methods (SFT, RLHF) tailored for clinical domains;

Develop and improve models that enable our AI agents to perceive and understand patients through video, language, and speech, building towards unified multimodal patient understanding;

Contribute to the full model development cycle: multimodal dataset curation and annotation, architecture design, cross-modal training strategies, evaluation, and iteration;

Collaborate across AI Engineering, Product, and Clinical teams to translate multimodal research breakthroughs into production systems that deliver patient care;

Work towards long-term ambitious research goals — such as real-time multimodal patient state estimation, clinical memory, and safety validation — while identifying and delivering immediate milestones;

Advance the field by publishing in top-tier AI venues and clinical journals, contributing to Sword's growing body of peer-reviewed research.

A PhD in Computer Science, Machine Learning, Natural Language Processing, Computer Vision, or a closely related AI field;

Hands-on experience fine-tuning large language models or multimodal large models (e.g., vision-language models, speech-language models), including pre-training, SFT, RLHF, or related post-training techniques;

Experience training or fine-tuning models that operate across multiple modalities (e.g., video + language, image + text, speech + text);

A strong publication track record in peer-reviewed AI conferences or journals;

Proficiency in Python and deep experience with modern ML frameworks (e.g., PyTorch, JAX);

Demonstrated ability to design rigorous experiments and interpret their results.

First-author publications in top-tier AI conferences (e.g., NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP, COLM, Interspeech);

Deep expertise in one or more of: vision-language models, video understanding, speech-language models, multimodal representation learning, or cross-modal fusion architectures;

Experience with video-based or image-based model training in applied settings (e.g., human pose estimation, action recognition, medical imaging, or biological signal processing);

Experience building or contributing to LLM-based agents, including prompt engineering, memory orchestration, or agentic workflows;

A track record of taking research ideas from conception to working systems, including developing and debugging complex multimodal ML pipelines;

Industry experience during or after the PhD (e.g., research internships at leading AI labs);

Comfort with ambiguity and a track record of delivering results in fast-moving, high-uncertainty environments where research and product development happen in parallel;

Strong communication skills and a history of effective cross-functional collaboration;

A broader record of research excellence demonstrated through grants, fellowships, patents, or impactful open-source contributions.