Toward Persistent and Reusable Reasoning in Large Language Models – Internship

Cambridge

AstraZeneca is a global, science-led biopharmaceutical business and its innovative medicines are used by millions of patients worldwide. AstraZeneca Summer Internships introduce you to the world of ground-breaking drug development, embedding you in highly dedicated teams, committed to delivering life-changing medicines to patients. Our 10–12-week program is designed for undergraduate, master's, and doctoral students. We offer exciting opportunities across Research & Development, Operations, and Enabling Units (Corporate functions).

Our internships immerse students in the pharmaceutical industry, allowing the opportunity to contribute to our diverse pipeline of medicines whether in the lab or outside of it. You will feel trusted and empowered to take on new challenges, but with all the help and guidance you need to succeed. This internship will help you develop essential skills, expand your knowledge, and build a network that will set you up for future success. You will be surrounded by curious, passionate, and open-minded professionals eager to learn and follow the science, fostering your growth in a truly collaborative and global team

Introduction to Role

Join our Data Science & AI team at AstraZeneca, where we are pushing the boundaries of AI-enabled drug discovery and clinical decision support. This internship sits within the Centre for AI, working at the intersection of foundation model research and production AI systems deplo

yed across our R&D pipeline.

You will work on a cutting-edge research project investigating how to make AI reasoning more efficient and reusable, a critical capability for long-running scientific analyses, clinical workflows, and multi-step drug discovery tasks. This work has direct applications to our agentic AI initiatives, where models must maintain reasoning context across multiple sessions, tools, and stakeholders.

The project explores novel approaches to managing reasoning states in large language models, enabling AI systems to checkpoint their reasoning progress and resume seamlessly. Your work will help determine how AI reasoning can be stored efficiently while preserving the ability to continue complex analytical workflows, and whether such approaches can work across different model architectures.

This is an opportunity to contribute to foundational AI research with immediate practical impact on how we deploy intelligent systems in healthcare and life sciences.

Accountabilities

As Part Of This Internship, You Will

Design and implement compression architectures that operate on latent representations from large language models, using attention-based mechanisms and information-theoretic principles

Conduct experiments to characterize efficiency-reliability trade-offs in reasoning systems, measuring how different compression strategies affect task performance

Evaluate approaches on reasoning benc

hmarks with verifiable outputs (e.g., mathematical, logical, or scientific reasoning tasks), focusing on functional capability preservation

Document findings through clear experimental reports, visualization of performance trade-offs, and recommendations for practical deployment scenarios

Collaborate with AI researchers to understand deployment requirements for persistent reasoning in production healthcare AI systems

Present results to stakeholders in Data Science & AI, translating technical findings into practical guidance for system design

Essential Criteria

Currently undertaking a PhD with focus in Machine Learning, Computer Science, AI, or related quantitative field

Experience with deep learning frameworks and implementing transformer-based architectures

Strong understanding of representation learning concepts and familiarity with attention mechanisms

Practical experience with training neural networks, including debugging, hyperparameter tuning, and experiment tracking

Programming proficiency in Python with experience in scientific computing libraries (NumPy, etc.)

Ability to read and implement methods from recent ML research papers

Strong analytical skills with ability to design experiments, interpret results, and identify patterns in model behavior

Desirable

Familiarity with information theory concepts (mutual information, compression principles)

Experience with large l

anguage models (fine-tuning, inference, or working with model internals)

Knowledge of latent variable models or variational methods

Publications or coursework in deep learning or NLP

AstraZeneca is where you can immerse yourself in groundbreaking work with real patient impact.

Trusted to work on important projects, you’ll have the independence to take on new challenges while receiving all the guidance you need to succeed. Our collaborative environment is designed to help you grow professionally and personally, surrounded by passionate individuals eager to make a difference.

Our mission is to build an inclusive and equitable environment. We want people to feel they belong at AstraZeneca, starting with the recruitment process. We welcome and consider applications from all qualified candidates, regardless of characteristics.

We offer reasonable adjustments/accommodations to help all candidates to perform at their best. If you have a need for any reasonable adjustments/accommodations, please complete the section in the application form.

Ready to make an impact? Apply now and join us on this exciting journey!

#Earlytalent

Date Posted

26-Jan-2026

Closing Date

09-Feb-2026

Our mission is to build an inclusive and equitable environment. We want people to feel they belong at AstraZeneca and Alexion, starting with our recruitment process. We welcome and consider applications from all qualified candidates, regardless of characteristics. We offer reasonable adjustments/accommodations to help all candidates to perform at their best. If you have a need for any adjustments/accommodations, please complete the section in the application form.