2027 Spring Intern - GenAI Applications

Roche•South San Francisco, CA

5d•$52•Onsite

About The Position

This internship position is located in South San Francisco, onsite. As an Applied GenAI Scientist Intern, you will explore how intelligent agents can reason, use specialized tools, and execute multi-step business workflows across Commercial, Medical, and Government Affairs (CMG). Your project will focus on developing and evaluating a simple, reusable approach for adapting advanced GenAI agents to business use cases. Building on the Deep Claude Agent approach, which uses Claude Code as an agent runtime, you will investigate how task instructions, reusable skills, context, and tool access affect agent performance. You will develop a working prototype and apply it to a focused CMG use case selected by your manager. Working with data scientists, business stakeholders, and platform partners, you will generate evidence and practical guidance for extending the approach to additional workflows.

Requirements

Must be pursuing a Master's Degree (enrolled student).
Must have attained a Master's Degree.
Must be pursuing a PhD (enrolled student).
Required Majors: Artificial Intelligence, Machine Learning, Computer Science, Data Science, Statistics, or a related quantitative discipline.
Foundational understanding of machine learning and generative AI, including large language models.
Experience using Python to conduct experiments, analyze data, or prototype AI applications through coursework, research, or personal projects.
Ability to design experiments, critically evaluate results, and communicate findings clearly.
Interest in applying GenAI to practical business problems and collaborating with domain experts.

Nice To Haves

Experience with LLM applications, AI agents, prompt design, or tool-augmented reasoning.
Familiarity with methods for evaluating LLM accuracy, reliability, and reproducibility.
Exposure to Claude Code, agent skills, or Model Context Protocol (MCP).
Experience fine-tuning, serving, or benchmarking open-weight models, including encoder-based classifiers.
Awareness of model licensing and data-handling terms when selecting third-party or open-weight components.
Interest in commercial, medical, or government affairs applications within healthcare or biotechnology.
Excellent communication, collaboration, and interpersonal skills.
A commitment to Integrity, Courage, and Passion in daily work and decisions.

Responsibilities

Define the use case and evaluation approach. Translate a selected CMG business need into a focused agent task, research questions, and measurable success criteria.
Design and prototype specialized GenAI agents. Combine task instructions, reusable skills, and relevant tools and data to enable an agent to carry out the selected workflow.
Assess and de-risk the model components the agent depends on. Compare hosted and open-weight options for specialized components such as the fast, calibrated classifiers used in routing, guardrails, and scoring, weighing accuracy, latency, and cost alongside licensing and data-handling terms. Recommend self-hostable alternatives where a dependency cannot be adopted as-is.
Conduct systematic experiments. Investigate how agent configuration, context, memory, and tool selection influence reasoning, output quality, and task completion. Analyze failure modes and identify where human review is needed.
Evaluate business value and reliability. Compare the agent’s performance with an agreed baseline, assessing accuracy, consistency, time savings, and execution cost.
Develop a reusable capability and share findings. Deliver a working prototype, adaptable agent configurations, and clear guidance that makes the approach easier to apply across CMG use cases. Present results, limitations, and recommendations to stakeholders.
Bring the latest techniques from academia into our organization and apply to our use cases.