Research Scientist Intern (Foundational Research)

Requirements:

• PhD student or recent graduate with research experience in a relevant discipline.
• Publications in top-tier conferences (e.g., NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL).
• Familiarity with a deep learning framework (e.g. PyTorch, JAX, TensorFlow).
• Excellent communication skills to report and present research findings and developments clearly, both orally and in writing.
• Curious and innovative disposition capable of devising novel, well-founded algorithmic solutions to relevant problems.
• Desirable: Experience working on at least one relevant state-of-the-art research topic in large language models (LLMs).
• Desirable: Influential first author publications in top-tier venues.
• Desirable: Impactful open-source contributions.
• Desirable: Strong software and/or infrastructure engineering skills with supporting evidence.
• Desirable: Experience training large-scale models over distributed nodes with cloud tools such as Amazon AWS, MS Azure, or Google Cloud.

What the job involves:

• We are seeking Research Scientist Interns with flexible starting dates throughout the year in our London, Toronto & Zug locations.
• During your internship, you will focus on publishing high-quality research in top venues for Machine Learning & NLP while advancing our internal model development.
• We also value our deep academic connections and are open to involving academic advisors & collaborators.
• Foundational Research is the dedicated core Machine Learning research division of Thomson Reuters, focused on research and development, particularly advanced algorithms and training techniques for Large Language Models (LLMs).
• We are building a strong foundation of research capabilities across different areas and are looking for interns who participate in designing, coding, conducting experiments, translating findings into concrete deliverables and engaging with the academic community.
• Our focus areas include:
• LLM Training (Continued Pretraining, Instruction Tuning, Reinforcement Learning Algorithms & Infrastructure, Alignment, Distributed Training).
• Post-training techniques for planning & reasoning (e.g. Agentic pipelines & tool use, LLMs & Knowledge Graphs, Self-reflection & critique, CoT & Reasoning, RAG).
• Data-centric Machine Learning (Synthetic & Hybrid Data generation, Curriculum Learning, learned data-mixtures).
• Evaluation (Benchmark design, Red-teaming/Adversarial Testing, Hallucination detection & Factuality, Human-in-the-loop testing).
• We work collaboratively with academic partners at world-leading research institutions and subject matter experts with decades of experience.
• We experiment, prototype, test, and deliver ideas in the pursuit of smarter and more valuable models trained on an unprecedented wealth of data and powered by state-of-the-art technical infrastructure.
• As a Research Scientist Intern, you will work alongside and learn from a diverse global team of experts.
• You will have the opportunity to publish your research findings as well as contribute to our proprietary AI model research & development.
• The internship duration at Thomson Reuters Labs is typically 4 to 6 months and may be aligned with one or two academic semesters or depending upon your availability.

In this opportunity, as a Research Scientist Intern you will:

• Innovate: You will have the opportunity to innovate and create new state-of-the-art ML/NLP/IR/GenAI approaches at the cutting edge of AI research.
• Experiment and Develop: You are involved in the entire research & model development lifecycle, brainstorming, coding, testing, and delivering high-quality reports at leading international academic conferences.
• Collaborate: Work on a collaborative global team of researchers & engineers both within Thomson Reuters and our academic partners at world-leading universities.
• Communicate: Actively engage in sharing our technical findings with the wider community through contributions to seminars, lectures, conferences and/or the sharing of publications and/or technical assets (data & models).