Senior Machine Learning Research Scientist

As a Machine Learning Research Scientist, you will lead groundbreaking ML research and development at SmarterDx, collaborating closely with experienced engineers and clinicians to turn your inventions into enterprise-grade products. With minimal supervision, you will establish a research agenda informed by emerging trends in AI, develop and rigorously evaluate your proposed algorithms, deploy your algorithms into production, and monitor impact. You will collaborate closely with other machine learning research scientists to identify the most promising areas of AI/ML R&D for SmarterDx and establish shared infrastructure to accelerate research efforts across the team.

• →45% Hands-on implementing new methods and relevant baseline models (ML Research)
• →20% Working cross-functionally to deploy models into production (MLOps, MLE)
• →20% Data Science (data engineering, dataset curation, experimental design, model updates, product domain expertise)
• →15% Academics & Outreach (e.g., scientific reading & writing, publishing, presenting at conferences, recruiting)
• →Become a domain expert at clinical data and the healthcare ecosystem
• →Own end to end model development including deployment into production and production monitoring, learning Machine Learning Operations (MLOps)
• →Post training to align large language models (LLMs) on proprietary clinical data
• →Develop new self-supervised pre-training tasks for improving models
• →Develop novel retrieval, attribution and hallucination detection strategies for generative models
• →Develop novel methods for explaining and summarizing diagnostic classifications
• →Develop methods for selecting data sources to include in training (data-centric AI)
• →Develop novel graph-based algorithms for improving classification of diseases and procedures with few or no labels
• →Develop novel methods for multimodal data fusion (structured and unstructured data)
• →Long-sequence language modeling

• Desire to translate research into tangible positive impact by deploying research into production engineering systems (MLOps)
• With scientific concepts, technical debugging or domain knowledge, ability and desire to communicate clearly and proactively when conveying or receiving
• Deep “under-the-hood” understanding of modern neural network architectures and distributed training. ie knows the differences between SwiGLU vs. sigmoid, GRUs vs. transformers vs SSMs, encoders vs. decoders, masked language models vs. autoregressive language models, Megatron vs nanotron vs DeepSpeed
• Extensive experience developing, implementing and training state-of-the-art deep learning models using multiple GPUs and nodes if necessary for large language models with frameworks such as PyTorch, JAX, etc
• Ability to assess, understand and create high-quality machine learning research, as demonstrated through publications at top-tier conferences and journals (e.g., NeurIPS, ICML, ICLR, ACL, EMNLP, SIGIR, AAAI, NEJM AI, JAMIA, npj Digital Medicine, arXiv)

• MLSys skills ie knows the differences between tensor vs pipeline vs data parallelism, gloo vs mpi vs nccl, CUDA vs ROCm, Triton vs ThunderKittens
• Familiarity with inference optimizations, ie vLLM, SGLang, continuous batching, KV Caching, speculative decoding

• PyTorch, Python, GitHub, Snowflake, Huggingface Transformers, AWS Sagemaker, Microsoft DeepSpeed, TorchTune, Apache Airflow, SLURM, Kubernetes