Staff Machine Learning Engineer

Responsibilities:

• Build, maintain, and improve supervised fine-tuning pipelines for open-source LLMs.
• Fine-tune models for healthcare administrative workflows involving multi-turn conversations, tool use, structured outputs, and task execution.
• Develop and refine training datasets from expert examples, workflow traces, synthetic data, user interactions, and model failure cases.
• Create evaluation frameworks to measure task completion, instruction following, factual grounding, tool-use accuracy, and regression quality.
• Analyze model failures and translate findings into improvements across data, training, prompting, retrieval, tooling, and product behavior.
• Collaborate with healthcare and RCM experts to learn workflows and convert domain knowledge into model behavior.
• Support production deployment, monitoring, and continuous model improvement.

About You:

• Hands-on experience fine-tuning open-source LLMs using supervised fine-tuning (SFT) pipelines.
• Strong Python and PyTorch skills.
• Experience with LLM tooling such as Hugging Face Transformers, PEFT, TRL, Axolotl, DeepSpeed, FSDP, vLLM, or similar frameworks.
• Experience preparing, cleaning, labeling, and validating instruction-tuning datasets.
• Familiarity with agentic LLM systems, including tool calling, structured generation, retrieval, or workflow execution.
• Experience evaluating LLMs using offline benchmarks, human review, regression testing, or production feedback.
• Strong debugging skills and the ability to systematically improve model behavior.

Preferred Qualifications

• Experience with RLHF, RLAIF, DPO, PPO, GRPO, reward modeling, or other preference-optimization methods.
• Experience deploying or monitoring LLMs in production.
• Experience with synthetic data generation, distillation, LoRA/QLoRA, or distributed training.
• Prior exposure to healthcare, fintech, insurance, operations automation, or other high-accuracy domains.
• Interest in learning healthcare revenue cycle workflows such as claims, denials, eligibility, prior authorization, and payer follow-up.

Who we are