Internship - AI Metadata Augmentation

About the role

Roche's Enterprise Metadata Management team is looking for a six-month internship to scale its AI metadata offering for multi-domain, enterprise-wide adoption. You will adapt and fine-tune AI agents to improve data discoverability, trust and compliance in a highly regulated environment, taking part in a project that moves directly into production under the guidance of experienced engineers.

Key responsibilities

• Scale and optimize AI agents across new business domains using prompt design, RAG and model fine-tuning.
• Build evaluation sets to measure agent performance on classification and quality checks, analyze failures and design confidence scoring and fallback logic.
• Prepare annotated datasets with subject-matter experts and integrate agent outputs into existing APIs, lightweight apps and enterprise workflows.
• Set up repeatable, tracked experiments with versioned datasets for reproducible improvement.
• Work in sprints with engineers and data stewards, presenting methods, limitations and demos to technical and business stakeholders.

Required profile

• Advanced Bachelor's or Master's student in AI, Data Science, Computer Science or a related field.
• Currently enrolled at a university, or completed studies within 12 months before the internship starts.
• Clear written and spoken English, intellectual curiosity, persistence and the ability to explain complex technical results to non-technical audiences.
• Non-EU/non-EFTA citizens must provide a university certificate stating the internship is mandatory and its required duration, plus a certificate of enrollment.

Required skills

• Strong Python, with hands-on experience in PyTorch/TensorFlow, classic ML tooling, SQL and Git/Docker/Linux.
• Experience building lightweight apps is a plus.
• Practical knowledge of the full ML lifecycle; preferred exposure to LLMs, agent frameworks and reward design.
• Familiarity with metadata, data catalogs or lineage is an advantage.