Machine Learning Scientist — Large Multimodal Models (Post-Training)

Iambic Therapeutics, Inc•Boston, MA

2d•Hybrid

About The Position

We are seeking a Machine Learning Scientist to join the Enchant team at Iambic Therapeutics. Our mission is to deliver better medicines through innovation in AI-based discovery technologies. In this role, you will research and develop post-training methods for Enchant - our multimodal transformer model trained on a wide variety of biomedical data - pushing the boundaries of what large-scale foundation models can achieve in drug discovery. The role centers on designing and evaluating post-training approaches for large multimodal language models including supervised fine-tuning, parameter-efficient fine-tuning, reinforcement learning, preference or reward-based optimization, and other emerging post-training methods. You will develop rigorous evaluations and training infrastructure that make it possible to rapidly iterate on these approaches at scale and work closely with colleagues across machine learning, software engineering, and drug discovery to put powerful foundation models into the hands of scientists making real therapeutic decisions. We are hiring across multiple levels and welcome candidates ranging from recent PhD graduates to experienced researchers with a strong publication or deployment record. This is a remote position, with the option to be on-site in our Boston office.

Requirements

PhD in machine learning, computer science, computational chemistry, physics, or a related computational STEM field, or equivalent industry experience demonstrating comparable depth
Strong Python and PyTorch skills, including implementing, training, debugging and evaluating deep learning models end-to-end
Demonstrated experience training large-scale transformer models
Demonstrated experience in one or more of the following: Reinforcement learning approaches such as RLHF, RLAIF, PPO, GRPO, RL with verifiable rewards, or related methods (strongly preferred)
Demonstrated experience in one or more of the following: Supervised fine-tuning, full-parameter fine-tuning, parameter-efficient fine-tuning (LoRA), or related methods
Demonstrated experience in one or more of the following: Systematic hyperparameter optimization or large-scale experimentation using tools such as Optuna, Ray Tune, or similar frameworks
Strong engineering practices: reproducible experimentation, clean code, testing, and performance-aware debugging
Comfort with modern ML infrastructure (e.g., Docker, CUDA, Kubernetes, experiment tracking tools such as Weights & Biases)

Nice To Haves

Experience with multimodal or multi-task model architectures
Training and inference optimization (e.g., mixed precision, kernel optimization, quantization, distributed strategies)
Familiarity with biomedical, chemical, or biological data domains
Distributed training at scale
HPC or large-scale training operations experience

Responsibilities

Research and develop post-training strategies for large-scale multimodal foundation models
Design reward functions, training objectives, data-generation strategies, and evaluation protocols for reinforcement learning and other post-training approaches applied to multimodal LLMs
Build systematic experimentation and hyperparameter optimization workflows to efficiently explore post-training recipes, model configurations, and training strategies
Develop and apply inference optimization techniques to support deployment in both high-throughput model evaluation and interactive discovery workflows
Design and maintain rigorous benchmarking and evaluation frameworks that measure model quality across modalities, downstream tasks, and scientific use cases
Collaborate with ML and software engineering colleagues to productionize models, evaluation systems, and inference services
Partner with computational chemists, medicinal chemists, and biologists to ensure model development and post-training objectives are grounded in drug discovery needs
Communicate results to internal teams, external partners, and at conferences
Write high-quality research and engineering code: refactor, test, document, and package ML components to support team velocity