Responsibilities
Formulate and investigate applied research questions in large language models, agentic AI, and machine learning for real-world spreadsheet scenarios.
Design and implement experiments, prototypes, and model or agent improvements, including prompt optimization, fine-tuning, post-training, retrieval, and tool-use approaches.
Build and curate representative datasets, benchmarks, and evaluation methods that measure task quality, reliability, latency, and efficiency.
Analyze model and agent behavior using quantitative metrics and qualitative error analysis; identify failure modes and translate findings into actionable improvements.
Collaborate with applied scientists, engineers, and product partners to integrate promising research ideas into our company 365 Copilot in Excel experiences.
Communicate methods, results, limitations, and recommendations clearly through technical documents, presentations, and peer reviews.
Requirements: You are pursuing your doctoral research under the supervision of an advisor who is actively conducting research in LLMs domain, generative AI, or closely related fields. Please include your advisors name in your CV or application.
You have at least one full year of study remaining before completing your degree and are available to dedicate a minimum of 2 full days per week to the internship.
Full proficiency in Python and common machine learning and deep learning frameworks such as PyTorch, TensorFlow, or equivalent tools. Good familiarity with using coding agents like Claude code and GitHub copilot.
Preferred Qualifications
Publications at recognized conferences or journals in the LLMs domain are highly valued, particularly work involving post-training, model adaptation, agentic AI, or evaluation. Please include references or links to relevant publications with your CV and briefly describe your contribution.
Demonstrated knowledge and experience in designing rigorous experiments, curating high-quality datasets, developing benchmarks and graders, and applying statistical methods to evaluate and analyze model quality and behavior.
Strong written and verbal communication skills and the ability to collaborate effectively in a multidisciplinary, globally distributed team.
Previous professional experience in software engineering is considered an advantage.
This position is open to all candidates.