Data Scientist Foundational Model Science

Position Summary

The Data Scientist for Foundational Model Science is the senior technical leader, and the lead scientist responsible for designing, training, and governing Mayo’s multimodal foundational model. This model forms the core intelligence layer used by clinical departments, researchers, agentic workflows, and sovereign AI collaborations. The individual will work as a hands-on architect, model-builder, and researcher while acting as a player–coach, guiding strategy and building a future team.

Key Responsibilities

Scientific & Technical Leadership

• Design multimodal foundational model architectures integrating signals from imaging, text, waveforms, structured data, graph representations, and temporal embeddings.
• Develop fusion, alignment, and cross-modal reasoning mechanisms (early fusion, late fusion, token-level fusion, hybrid models).
• Define and implement methods for grounded clinical reasoning, retrieval-augmented inference, graph-augmented attention, and chain-of-thought verification.
• Establish protocols for model lifecycle governance, safe update cycles, drift-aware re-training, and provenance tracking.

Hands-On Modeling & Training

• Train large-scale deep learning models, including multimodal architectures and domain-specific transformer-based systems, on real clinical datasets.
• Fine-tune and adapt large language models (LLMs) for clinical reasoning, summarization, question answering, agentic behavior, and instruction-following tasks.
• Build retrieval-augmented pipelines using embeddings, vector stores, graph traversal, and clinically grounded context construction.
• Develop evaluation methods for reasoning quality, temporal prediction accuracy, multimodal synergy, ablation-based robustness, and counterfactual behavior.
• Create reference-grounded training datasets, structured reasoning tasks, and multimodal benchmarks to evaluate model performance.
• Conduct hands-on experimentation with optimization strategies, large-scale distributed training, model quantization, and inference acceleration.
• Implement uncertainty modeling, selective prediction, abstention mechanisms, and clinically meaningful risk thresholds.
• Build interpretable reasoning pathways, cross-modal attribution maps, and reference-grounded explanations.

Cross-functional Collaboration

• Work closely with the Representation team to ensure representation-model alignment.
• Partner with clinical SMEs to encode domain reasoning into reinforcement learning, preference optimization, or rule-guided behaviors.

Team Leadership

• Serve as the future founding technical lead of the Foundational Model Science Program.
• Mentor scientists and engineers and eventually build a specialty modeling team.

Required

• PhD in Machine Learning, Computer Science, Applied Mathematics, or related discipline with at least four years of informatics, Artificial Intelligence, data science and/or machine learning.
• Experience with generative modeling, reasoning models, or multimodal foundation models.
• Expertise in alignment methods (contrastive learning, RLHF/RLCS, preference optimization).
• Experience with distributed training, and large-scale compute.

Preferred

• Experience with clinical or EMR data across multiple modalities.
• 7+ years experience training deep learning models, including transformers or multimodal architectures.
• Experience defining evaluation frameworks for reasoning, multimodal synergy, reliability, or fairness.
• Publications in multimodal learning, foundation models, or reasoning architectures.