Senior Applied Scientist, Generative AI

Liberty Mutual Fire Insurance Co.; Boston (MA); Hybride; USD 150 000 - 190 000; Plein temps

Résumé du poste

Liberty Mutual Fire Insurance Co. is seeking an experienced GenAI Research Scientist to lead end-to-end language model projects. You will fine-tune models, build retrieval-augmented pipelines, and collaborate with engineering to productionize solutions on Kubernetes.

You will distill models for domain-specific insurance tasks, design evaluation suites, and drive reproducibility through MLOps best practices while communicating results to diverse stakeholders.

Qualifications

• PhD with 2+ years or MS with 4+ years, or BA with 6+ years in a quantitative field.
• Expertise with transformer models and fine-tuning techniques like LoRA/QLoRA.
• Hands-on GenAI pipelines design including retrieval design and embedding strategies.
• Strong Python, PyTorch, ML tooling and evaluation skills.

Responsabilités

• Design, fine-tune and deploy language-model solutions end-to-end.
• Build GenAI pipelines with RAG, agentic orchestration and tools.
• Develop scalable data, embedding, and MLOps pipelines for reproducibility.
• Evaluate model groundedness, safety, latency and cost in production.

Connaissances

Transformer models

LoRA/QLoRA

Python

PyTorch

GenAI pipelines

Kubernetes

Docker

MLOps / LLMOps

Git / GitHub

Vector search

Formation

Ph.D. in Statistics / CS / Math / Econ / Actuarial Science

Outils

Hugging Face

LangChain

LlamaIndex

LangGraph

MLflow

Git

Airflow

Kubernetes

Helm

Description du poste

At GenAI Research and Solution (part of deep learning research team within modeling sophistication, DSE), we research and build generative AI (GenAI) capabilities that go directly into Liberty Mutual products. We fine-tune small language models (SLMs), design retrieval and agentic pipelines, explore creative ways to embed Liberty's data in products, and hold all these solutions to a high bar for accuracy, grounding, latency and cost. Our team emphasizes technical rigor, reproducibility and methodological innovation, and we work close to the business so that research turns into products our customers use.

As an individual contributor on the team, you will provide technical leadership, design, build and deploy GenAI solutions end to end - from research prototype to production service. You will fine-tune and evaluate small language models, build pipelines such as retrieval-augmented generation (RAG), and variants such as corrective RAG and agentic orchestration, and partner with engineers to run them reliably on our Kubernetes and other deployment platforms. This is a deeply hands‑on role with room to shape methodology and influence how GenAI shows up across our products.

Candidates who live within 50 miles of Boston, MA; Portsmouth, NH; Seattle, WA; Columbus, OH; or Plano, TX will follow a hybrid schedule, coming into the office two days per week. Otherwise, this role is remote with occasional travel.

Responsibilities:

• Design, fine‑tune and deploy language model based solutions, from research and experimentation through production implementation.
• Fine‑tune and distill small language models for domain‑specific insurance tasks, balancing accuracy, latency and cost.
• Build and improve GenAI pipelines, including RAG, corrective RAG and agentic orchestration with tool use, planning loops and memory.
• Develop and maintain scalable data, document and embedding pipelines, applying MLOps and LLMOps best practices for reproducibility, deployment and monitoring.
• Design evaluation suites and guardrails for GenAI use cases, covering groundedness, accuracy, safety and regression testing as models and prompts change.
• Containerize and ship solutions on Kubernetes, partnering with engineering teams to operationalize them in production environments.
• Research and prototype new methods for training, adapting and evaluating generative models, and share what works with the wider team.
• Communicate findings through technical presentations, reports and recommendations to both technical and non-technical stakeholders.
• Participate in cross-functional working groups and contribute to the broader data science community to promote best practices.

Preferred qualifications:

• Ph.D. in Statistics, Computer Science, Mathematics, Economics, Actuarial Science or a related quantitative field with 2+ years of relevant experience; or Master's with 4+ years; or Bachelor's with 6+ years.
• Demonstrated expertise with transformer-based language models, including fine-tuning techniques such as parameter-efficient fine-tuning (LoRA/QLoRA) and instruction tuning.
• Hands‑on experience building GenAI pipelines, including retrieval design, chunking and embedding strategies, vector search and agentic tool use.
• Strong foundation in machine learning, statistics, experimental design and model evaluation metrics, including the evaluation of generative output.
• Proficiency in Python and MLOps practices, with experience in version control (Git), code review, collaborative development workflows (e.g., GitHub/GitLab) and model versioning/experiment tracking (e.g., MLflow).
• Proficiency in PyTorch and the GenAI ecosystem, such as Hugging Face, LangChain, LlamaIndex or LangGraph.
• Experience building and managing pipelines with workflow orchestration tools (e.g., Airflow, Luigi).
• Ability to clearly communicate technical concepts to diverse audiences.
• Track record of advancing research projects from ideation to implementation.
• Experience with Docker and CI/CD pipelines.
• Experience deploying and scaling containerized workloads with Kubernetes, including Helm and GPU‑backed services.
• Understanding of GPU acceleration, distributed training and inference optimization techniques (e.g., mixed precision, quantization, KV caching, request batching).
• Experience with multimodal models and cross‑modal retrieval, including vision‑language models.
• Experience with insurance data, or with regulated‑industry constraints such as responsible AI review, data governance and auditability.

Qualifications

Broad knowledge of predictive analytic techniques and statistical diagnostics of models.

Expert knowledge of predictive toolset; reflects as expert resource for tool development.

Demonstrated ability to exchange ideas and convey complex information clearly and concisely.

Networks with key contacts outside own area of expertise. Ability to establish and build relationships within the aligned functional area or SBU.

Ability to give effective training and p