EPSRC Summer Internships 2026

Supervisor: Akshat Mudgal

Department: Mathematics Institute

Let G be an abelian group, and let A_1,...,A_k be finite sets in G. One is often interested in studying solutions to symmetric linear equations of the form x_1 + ... + x_k = y_1 + ... + y_k, with x_i,y_i being elements of A_i. We call this number of solutions E(A_1,...,A_k). It is natural to ask whether this inhomogeneous count E(A_1,...,A_k) can be controlled by its homogeneous variants E(A_1,...,A_1), ..., E(A_k,...,A_k). When G is an abelian group, this can be done via Fourier analysis and Holder's inequality. These methods do not extend to the non-abelian setting. In this project, we are interested in studying and proving such inequalities without using Fourier methods and relying instead on purely combinatorial techniques. This would help us generalise these results to the setting of non-abelian groups. This problem already exhibits interesting connections to linear optimisation type problems, so this would also be a nice relationship to explore.

Supervisor: Bora Karasulu

Intern: Thanniya Sorubanathan

Department: Chemistry

Boron clusters are small nanoscale assemblies of boron atoms that can adopt intricate cage-like and polyhedral structures. These unusual geometries give rise to distinctive electronic and bonding properties, making boron-rich clusters scientifically interesting as model nanomaterials and as possible building blocks for future functional chemical systems. However, predicting which structures are most stable, and how their properties change with geometry, is computationally demanding. Density functional theory (DFT) is a well-established and reliable quantum-mechanical method for studying such systems, but its high computational cost limits the number of candidate structures that can be explored. This project therefore asks whether modern machine-learning interatomic potential models can reproduce the key structural and energetic features of boron clusters at much lower computational cost, thereby enabling faster and broader computational screening. This sits clearly within EPSRC remit through its focus on computational chemistry, atomistic simulation, and AI-enabled scientific modelling, and it is well suited to the scheme’s requirement for a clearly defined project that is feasible within an 8-week full-time internship.

The internship will investigate a defined set of boron cluster structures using a staged computational workflow. First, the student will undertake a short literature review to understand the main structural motifs of boron nanoclusters and the current approaches used to model them. Candidate structures will then be generated using Ab Initio Random Structure Search (AIRSS). These structures will be optimised using selected machine-learning interatomic potential models implemented through established Python-based atomistic simulation workflows. A representative subset will then be benchmarked using DFT calculations, providing a reference against which the machine-learning predictions can be evaluated. The student will compare the methods in terms of predicted minimum-energy structures, retention of key topological features, computational efficiency, throughput, and practical limitations. This overall design follows a proven and accessible project model already used successfully within the group for related machine-learning-versus-DFT benchmarking studies on nanocluster systems.

The project is deliberately scoped to keep it suitable for an undergraduate researcher within 8 weeks. Week 1 will focus on induction, background reading, and training in the software environment and computational workflow. Weeks 2–3 will cover structure generation and initial screening. Weeks 3–6 will focus on machine-learning-based optimisation and data collation. Weeks 5–7 will involve selected DFT benchmarking calculations and comparative analysis. Weeks 7-8 will be used to consolidate the results and prepare a final report and poster. The project will be carried out using Warwick’s high-performance computing facilities, with the student based within the group’s computational research environment and supported throughout by regular supervision. Expected outputs are a curated dataset of boron cluster structures, a comparative benchmark between machine-learning and DFT approaches, and a short written and visual summary of the scientific findings.

Supervisor: Hungyen Lin

Intern: David Humphries

Department: School of Engineering

Ion‑exchange membranes are crucial components in electrochemical energy‑conversion devices such as fuel cells and electrolysers, yet achieving an optimal balance between performance, durability and manufacturability remains a significant challenge. Because water plays a central role in determining membrane properties, there is strong strategic need for rapid, non‑destructive methods that can characterise water properties and transport with high fidelity. Our group has recently demonstrated such a capability through emerging humidity‑controlled terahertz time‑domain spectroscopy, which allows detailed water characterisation under humidity controlled conditions [1–3]. This internship will support an ongoing EPSRC project by further developing and optimising this capability, and by integrating it with advanced instrumentation and custom software to unlock new functionalities for routine membrane metrology.

Plan of activities include:

1. CAD design and realise a fully environmentally controlled chamber with appropriate dimensions using quartz windows and available parts in the laboratory (e.g. mass flow controllers, Pythons for proportional–integral–derivative control) for integration with the terahertz system;

2. Benchmark the operational performances of the chamber (dynamic response, accuracy and precision) at discrete levels of relative humidities at various temperatures and fine-tune controller parameters;

3. Deploy the chamber and acquire membrane water data at discrete levels of relative humidities for benchmarking against complementary techniques e.g. NMR, dynamic vapour sorption.

Strategically, this work is aligned to EPSRC remit of 1) Manufacturing the Future by providing routine and accessible metrology relevant to membrane production and optimisation; 2) Energy and Decarbonisation by improving understanding of water uptake and transport in materials central to energy technologies.

By enabling detailed probing of membrane water behaviour under full environmental control, this project will allow comprehensive comparison of terahertz measurements with state‑of‑the‑art methods such as differential scanning calorimetry and nuclear magnetic resonance across a full range of conditions. This will provide the foundation for systematic exploration of membrane parameter space, informing future material design and optimisation. We anticipate to disseminate research outputs in target journals include J. Mater. Chem. A, Energy Environ. Sci., Nature Mater., J. Electrochem. Soc., IEEE Trans. Terahertz Sci. Technol. and Opt. Express. This work contributes directly to an awarded EPSRC grant EP/Z534237/2 and is supported by strong industrial and academic collaboration. Johnson Matthey, a leading UK PLC, and internationally recognised membrane researchers at the University of Surrey will provide membrane samples free of charge, ensuring the internship has real‑world impact and relevance.

[1] S. A. Franklin et al. Journal of Materials Chemistry A, 2026.

[2] G. A. H. France et al. IEEE Transactions on Terahertz Science and Technology 15, 743, 2025.

[3] G. A. H. Ludlam et al., ACS Sustainable Chemistry & Engineering 12, 7924–7934, 2024.

Supervisor: Mark Greenhalgh

Intern: Amandeep Bhairon

Department: Chemistry

The ability to control the structure of molecules is essential for the design of new pharmaceutical agents, where the 3D structure of these molecules can have a pronounced effect on biological activity. The ability to control the 3D structure of molecules through a catalytic process is commonly referred to as ‘enantioselective catalysis’. Progress in this field has traditionally been accomplished using transition metal catalysis and organocatalysis and has been recognised by two recent Nobel Prizes in Chemistry (2001, 2021).

Chalcogen bonding is an unconventional non-covalent interaction that arises between electrophilic chalcogen-containing molecules (S, Se, Te) and Lewis bases and is being increasingly applied in catalysis and supramolecular chemistry. Recent work in the Furata and Greenhalgh groups has shown that intramolecular chalcogen bonding can be used to provide a conformational lock and increase the hydrogen bond donor ability of benzoselenophene-substituted thioureas. Recent unpublished work in the Greenhalgh group has demonstrated these thioureas are highly active hydrogen bonding catalysts. In this project, chiral variants of these catalysts will be synthesized and tested as novel enantioselective hydrogen bonding catalysts.

Building on computational work that Maria (the named student) undertook with collaborators at VUB last summer (EUTOPIA-funded project), she will synthesise a small range of novel chiral hydrogen bonding catalysts that incorporate a benzoselenophene substituent, and assess their activity and enantioselectivity against current state-of-the-art literature catalysts. This will involve the synthesis and characterisation of novel compounds, followed by kinetics and chiral HPLC analysis to assess the activity and enantioselectivity of the new catalysts. This initial comparison and reaction optimisation phase will be completed within the timeframe of the project, providing the basis to explore the substrate scope of the developed reactions either by Maria or an MChem student joining the group in October. This work will result in publication in an international peer-reviewed journal within 12 months.

The project fits within the EPSRC 'Physical Sciences' portfolio and aligns with the strategic priority on 'the physical and mathematical sciences powerhouse' and 'transforming health and healthcare'.

Supervisor: Frederik Schaufelberger

Intern: Riley Canning

Department: Chemistry

Introduction and project aims:

Mechanically interlocked molecules (MIMs) consist of discrete components connected via mechanical bonds, linkages that allow independent movement of components but prevent their mutual dissociation (10.1002/9781119044123). For example, rotaxanes consist of a flexible thread encircled by a macrocycle, which is prevented from slipping off the thread by bulky stoppers situated at each end. The interlocked nature of MIMs leads to “steric shielding”, meaning that the mechanical bond protects active sites on the thread. We can use this feature for so called “mechanical silencing” of function – a thread can be sterically protected with a macrocycle until a certain stimulus is applied, which cleaves the macrocycle and liberates the active molecule.

A good candidate for this chemistry is peptides. These are fundamental units in biology as the main constituent of proteins, finding ubiquitous use in therapeutics and diagnostics. The global peptide therapeutic market is worth in excess of 50 billion EUR, but the low metabolic stability of these units is an issue preventing wider applicability. Our team is currently making mechanically interlocked peptides (MIPs) where the mechanical bond is used to stabilise the bioactive units (10.1021/jacsau.5c00794). We use the well-known propensity of primary amines to undergo condensation reaction through the cavity of a 24-crown-8 (24C8) macrocycle with activated esters to form interlocked amide bonds (10.1002/syst.202300048). This crown ether active template method can be used to turn Gly-containing bioactive peptide sequences into rotaxanes. Subsequently, using stimuli-responsive macrocycles which can “self-immolate” on-demand, we release the bioactive peptide at targeted sites in the body (10.1002/anie.202400344). However, synthesis of these bespoke responsive macrocycles has been very labour-intensive and difficult, taking months of lab work to obtain testable quantities of even a single cleavable macrocycle.

To solve this synthesis issue, we are developing new approaches to cleavable macrocycles using photoredox catalysis (10.1021/acs.joc.6b01449). In this internship, we will evaluate a one-pot functionalisation method where we use a photocatalyst to functionalise a macrocycle with a halogen atom, and then diversify this functionalised intermediate with cleavable groups. The project is focused on synthetic organic and supramolecular chemistry and hence falls perfectly within the EPSRC remit.

Plan of activities:

During the internship, the student will work on their own project line under the direct guidance of Dr. Schaufelberger and a postdoc in the group. The student will optimise the new photoredox method using a model 18-crown-6 macrocycle and screen halogenating agents, photocatalysts and photocatalytic setups versus reaction outcome. Next, the student will explore different which cleavable stimuli can be attached, and if time permit the scope will be expanded to making peptide rotaxanes with all these new cleavable macrocycles to fully test the concept.

All compounds will be synthesized via multistep organic synthesis, purified and characterised (NMR, HRMS, HPLC, FTIR, etc) after each subsequent reaction step.

Proposed output:

The results from the internship will be published as part of a high-impact research paper describing our work on photoredox-catalysis in rotaxane synthesis. The student will also present the results at the annual symposium for undergraduate chemistry research at UoW.

Supervisor: Feng Hao

Intern: Darsh Kothari

Department: Computer Science

As electoral voting becomes increasingly computer-assisted, end-to-end (E2E) verifiable voting systems use cryptographic techniques to provide verifiability without sacrificing voter privacy. Alongside E2E systems, risk-limiting audits (RLAs) are statistical post-election methods that guarantee a reported outcome is correct by examining a randomly selected sample of ballots. While RLAs are efficient, making them publicly verifiable traditionally risks voter privacy.

This project explores how concepts from end-to-end (E2E) verifiable voting can be applied to Risk Limiting Auditing (RLAs) to allow the public to gain confidence in election outcomes using public data. The primary objective for the intern is to develop a proof-of-concept prototype, implemented in Python, to test the feasibility of this publicly verifiable RLA scheme. A secondary aim is to contribute to the statistical analysis. They will simulate the effect of varying parameters, such as risk thresholds or margins of victory, on the security guarantees of the RLA scheme.

This project aligns directly with the EPSRC remit within Information and Communication Technologies, as well as the Mathematical Sciences. With the completion of this project, the team would be closer to designing a framework that would lead to a more equitable digital society. Access to credible data can enhance voter empowerment, thereby increasing engagement in the electoral process and helping to address societal needs.

The advancement of RLA technologies would clothe electoral polls in much less ambiguity and lead to significant increases in public confidence following elections. There is still a long wait before this technology becomes widely adopted; however, it is seeing some real use cases in other parts of the world, and advancement in this project would only help it gain traction.

The project will begin with a review of current literature to strengthen the intern’s understanding of both related concepts and the theory for which they will help provide proof. Following this, the intern will theoretically design the program to detail the necessary parameters and output formats. Once the theoretical framework has been verified, they will implement the proof-of-concept program in Python, consulting technical documentation to expand their capabilities. The intern will then rigorously test the code using open-source government datasets, as well as theoretical edge cases such as extreme margins of victory. After ensuring that the program runs as expected, the intern will run the provided research data through the program, methodically organising the results. Finally, the intern will conclude by collating the results into an appropriate document for specialist team members, highlighting key results.

The proposed outputs are a functional Python prototype and a comprehensive document collating results with clear statistical visualisations for the research team. Additionally, the intern will produce an accessible poster targeting a non-specialist audience for the URSS showcase in November.

Supervisor: Sebastian Pike

Intern: Hamza Iqbal

Department: Chemistry

This project will synthesize new ultrasmall copper oxide nanomaterials, with unique properties that will go on to be tested as photocatalysts for the production of green hydrogen.

New advanced materials for sustainable processes sits clearly within the EPSRC remit.

Developing simple Earth-abundant semiconductor materials to harvest sunlight and convert this energy into green H2 at scale is a crucial scientific challenge that requires urgent advancement. Global H2 demand will increase by 1.5 times by 2030, yet only 1% is currently ‘low-emission H2’. Copper(II) oxide (CuO) is an Earth-abundant small band gap semiconductor which could fulfil this role, however, as a bulk material its electronic structure is not suitable for the oxidation of water. Recent discoveries indicate that ultrasmall (<3 nm) nanoparticles (NPs) of CuO benefit from quantum size effects that increase the oxidation potential of photochemically generated holes, which enables unassisted photochemical water oxidation (to make H2). This exciting discovery implies that ultrasmall CuO could become a key new material for water splitting devices.

Nanomaterials have revolutionised many fields including electronic components, catalysis, health & medicine and self-cleaning materials. Synthetic understanding has provided enormous control of size, shape and composition of nanomaterials allowing customisable properties to suit specific applications. There is continued interest in even smaller nanoparticles (NPs), with the smallest termed ‘ultrasmall’, defined as ≤3 nm in size and typically with less than <1000 atoms. Remarkably the electronic, surface and structural properties of NPs at these dimensions can be drastically altered, making them truly unique materials with their own specific properties.

The synthesis of these small particles is challenging, whilst introducing dopants or accessing ternary phases is very underexplored due to practical difficulties. This project will explore transforming metal alkoxide, metal carboxylate precursors using mild conditions in the solid state to make soluble ultrasmall nanoparticles.

The project will focus on the synthesis of CuO nanoparticles and doped variants. Recent results from our lab have shown that CuO has very different properties at the ultrasmall scale. The project will expand on recent Pike group research by using new precursors such as copper methoxide in a solid-state route to ultrasmall particles. The student will study the synthetic process, exploring variables such as temperature, reaction time, ligand choice and concentration. The particles will be characterised by powder X-ray diffraction, UV and IR spectroscopies, small-angle X-ray scattering and transmission electron microscopy analysis, working with experienced members of the Pike group and RTP technical professionals to achieve these analytical processes, and this will provide the student with a wealth of new experiences and understanding of important techniques.

The student will also explore introducing dopants through a simple mixing strategy or a more advanced single-source precursor approach where multiple metals are combined in one molecular precursor.

The enhanced understanding of synthetic routes will be of significant value to future research, and will contribute to publications.

Supervisor: Shan Raza

Intern: Demir Bibezic

Department: Computer Sciences

Project Aims and Motivation:

TIAToolbox (https://github.com/TissueImageAnalytics/tiatoolbox) is an open‑source Python library that provides state‑of‑the‑art methods for whole slide image handling, visualisation, and deep learning in computational pathology with nearly 300K downloads. While the toolbox offers rich functionality through programmatic interfaces, interaction with its visualisation and analysis capabilities currently requires familiarity with Python and command‑line driven workflows. This can present a barrier for non‑programming users e.g., pathologists and interdisciplinary collaborators.

The aim of this project is to design and evaluate a Napari‑based standalone viewer for TIAToolbox, enabling intuitive visualisation of supported image formats and interactive deployment of selected built‑in AI models through a graphical user interface supporting research and development in computational pathology. The project will critically investigate design choices, usability, and performance trade‑offs involved in integrating machine learning workflows into an interactive desktop application, providing the intern with a genuine research experience.

Fit to EPSRC Remit:

The project aligns with the EPSRC remit through its emphasis on:

• Software engineering and research software sustainability
• Machine learning system integration and model deployment
• Computational methods and algorithmic workflows to promote interdisciplinary research

The research focus is on methodological and engineering challenges, not biomedical interpretation. The outcomes relate to generalisable principles for making advanced AI tools accessible through robust software design. The intern will have the opportunity to work closely with our clinical collaborators who can provide feedback on the interface utility.

Plan of Activities:

Weeks 1–2: Background and System Design

The intern will be introduced to TIAToolbox architecture, Napari’s plugin framework, and best practices in research software engineering. A short literature and technical review will inform design requirements, followed by an initial viewer architecture.

Weeks 3–4: Core Visualisation Functionality

Implementation will focus on loading and visualising image formats supported by TIAToolbox, including navigation, zooming, and layer management within Napari. Design decisions will be documented and justified.

Weeks 5–6: AI Model Integration and Interaction

Selected pre‑trained deep learning models within TIAToolbox will be integrated into the viewer. The intern will design a graphical interface for model selection and parameterisation, enabling model inference directly from the viewer and visualisation of outputs such as segmentation masks or heatmaps.

Week 7: Evaluation and Refinement

The intern will evaluate usability, responsiveness, and robustness of the viewer through structured testing with members of the research group including our clinical collaborators. Limitations and potential improvements will be critically analysed.

Week 8: Dissemination and Reflection

The project will conclude with documentation, a final technical report, and a presentation to the host research group and the UG cohort, reflecting on research findings and broader implications.

Proposed Outputs:

• A functional Napari‑based standalone viewer for TIAToolbox
• Well‑documented code suitable for future extension
• A written research report analysing design and evaluation outcomes
• A presentation demonstrating the software and research insights

Overall, the project provides an achievable yet intellectually rich research experience, exposing the undergraduate intern to the full research cycle while contributing meaningfully to an active software research ecosystem.

Supervisor: Rianne Lord

Intern: Leilani Bellamy

Department: Chemistry

Project Aims and Motivation:

Platinum-based chemotherapeutics, such as cisplatin, are used in approximately 40–50% of cancer treatments and demonstrate high efficacy in certain cancers. However, their clinical application is limited by severe drawbacks, including toxicity, tumour resistance, poor selectivity, and the high cost and scarcity of platinum. These limitations motivate the development of alternative metal-based drugs that are more selective, less toxic, and economically viable.

This project focuses on vanadium as a promising alternative metal centre for anticancer therapeutics. Vanadium is a low-cost, earth-abundant, and biologically relevant metal, offering potential advantages in reducing patient side effects. Previous work by the Lord research group has demonstrated that vanadium complexes exhibit strong cytotoxicity toward cancer cells while maintaining high selectivity over healthy cells. Proposed mechanisms include interactions with double-stranded DNA and the generation of reactive oxygen species, leading to cancer cell death. However, the intracellular localisation and precise mechanisms of action of these compounds remain poorly understood.

The primary aim of this project is to develop traceable vanadium-based drug candidates to enable direct investigation of their behaviour within cancer cells. This will be achieved through the design and synthesis of organic fluorescent tags, which will be coordinated to vanadium centres. These fluorescent complexes will allow real-time visualisation of drug uptake and intracellular distribution.

The research plan includes 3 main objectives and the following timescale:

1. Synthesis and characterisation of a fluorescent organic ligand (weeks 1-3)

2. Synthesis and characterisation of coordination vanadium complexes which incorporate the fluorescent tag (weeks 3-6)

3. Biophysical measurements to understand the fluorescence lifetime, hydrophobicity (for cell uptake) and stability (to determine drug incubation times) of the vanadium complexes.

Additional biological testing will be carried out in collaboration with Lord group members to evaluate cytotoxicity against mammalian cancer cell lines, whilst fluorescence and confocal microscopy will be used to determine intracellular localisation, providing new insight into uptake pathways and mechanisms of action.

This project aligns with the EPSRC remit by advancing fundamental understanding in inorganic and bioinorganic chemistry, while integrating molecular design with imaging technologies. It addresses key challenges in drug mechanism elucidation and targeted delivery. Expected outputs include a series of fluorescently labelled vanadium complexes, enhanced understanding of their biological activity, and dissemination through publications and presentations, contributing to the development of next-generation metal-based anticancer therapies.

Supervisor: Rosalie Thompson

Intern: Maria Gonsalves

Department: Physics

In this project, we aim to develop and use solid-state NMR approach to study the structure and interactions of the different forms of mannan within primary cell walls. In the primary cell walls, there is Mannan which is a polymer of mannose; Glucomannan which is a polymer of glucose and mannose; and galactoglucomannan which is glucomannan with sidechains of galactose. It is thought that these different polymers play different roles within the primary cell wall. Therefore, their structure and how they interact with other components, such as cellulose, will help identify their roles.

Recent work by the group has identified the structure of secondary cell walls with a particular focus on cellulose and xylan. The difficulty in this project is that the chemical shifts of mannan, glucomannan and galactoglucomannan are very similar and so we must test different solid-state NMR experiments to resolve the structures of these three polymers. Paul Dupree’s group at the University of Cambridge will provide 13C-labelled primary cell wall samples enabling the acquisition of a range 2D NMR spectra. There will also be access to samples with genetic modifications to remove one of the mannan polymers. This would aid with assignment of chemical shifts and help identify how the cell wall changes in the absence of this polymer.

This project fits well into EPSRC’s remit as we aim to develop the solid-state NMR method to resolve the mannan structure. Solid-state NMR is not a standard technique biochemists use in the plant cell wall community as it requires a good grasp of the quantum physics behind the method to acquire and analyse NMR spectra effectively. The student will be a physics undergraduate who has a good grasp of the underlying quantum theory of NMR.

The student would initially learn to set up and acquire NMR spectra on a couple of 13C-labelled samples. We will discuss and trail a range of 2D NMR experiments to identify what set of experiments are ideal for studying the mannan structure and interactions. (Week 1-3)

The student will then be introduced to and learn how to analyse the NMR spectra using Topspin. At first the student will focus on assigning the different mannan polymers and we will discuss what any differences in chemical shift could mean for their structure within the cell wall. (Week 3-5)

Once assignments have been made the focus will shift to comparing the dynamics of the different types of mannan and if they significantly interact with other polymers in the cell wall. It will be interesting to see if there is a particular mannan polymer that interacts with celluloses whilst others are more hydrated. (Weeks 4-7)

Finally, the student will collate their results in a presentation where they will present at a collaborator meeting at the University of Cambridge. At this meeting we will discuss the results and decide how this fits with a publication plan. The student has also registered to present their results at the International Conference of Undergraduate Research (ICUR) in September. (Week 7-8)

Supervisor: Shan Raza

Intern: Hannah Wood

Department: Computer Sciences

Modern computational pathology relies on the analysis of extremely large digital histology images, often comprising billions of pixels. While contemporary machine-learning frameworks are highly optimised for model inference, many upstream stages of the analysis pipeline such as whole-slide image handling and processing remain performance bottlenecks when implemented in Python. These stages are typically CPU-bound, data-intensive, and executed repeatedly, limiting throughput and scalability.

This project aims to evaluate Rust as a high-performance complement to Scientific Python in computational pathology, focusing on accelerating performance-critical components of existing Python-based workflows rather than replacing them. Rust has recently emerged as a compelling systems-level language for scientific and developer tooling, demonstrating substantial performance gains in real-world Python-adjacent projects. Examples such as Ruff (a Python linter and formatter) and uv (a Python package and environment manager) show that Rust-based implementations of performance-critical functionality can achieve multiple-times to order-of-magnitude speed improvements compared to traditional Python-based tools, while maintaining seamless integration with the Python ecosystem. This project will investigate whether similar benefits can be realised in the context of computational pathology workflows.

The project sits firmly within the EPSRC remit, addressing research challenges in software engineering, programming languages, high-performance and parallel computing, and data-intensive scientific software infrastructure. The emphasis is on computational methods and research software design principles, rather than biomedical experimentation, with computational pathology serving as a demanding and realistic application domain.

Project Plan and Activities

Weeks 1–2: Background and Profiling The intern will be introduced to computational pathology pipelines and an existing Python-based image analysis workflow. Profiling tools will be used to identify performance-critical components (e.g. patch extraction or image normalisation) that are suitable for re-implementation in Rust. Evaluation metrics such as runtime, memory usage, and scalability will be defined.

Weeks 3–5: Rust Implementation and Python Integration One carefully scoped component will be re-implemented in Rust, leveraging Rust’s strengths in memory safety and parallel execution. The component will be exposed to Python via established interoperability tooling, allowing it to act as a drop-in complement within an existing Scientific

Python workflow. Correctness will be validated through comparison with the original Python implementation.

Week 6: Optimisation and Workflow Integration The Rust component will be optimised and integrated into a representative end-to-end computational pathology workflow. Where appropriate, parallelisation strategies will be explored to better utilise multi-core CPUs.

Week 7: Evaluation and Benchmarking Quantitative benchmarks will compare the Rust-based component with the original Python version, evaluating performance improvements, memory behaviour, and developer usability, alongside any trade-offs.

Week 8: Reporting and Dissemination The intern will produce a concise technical report and reproducible benchmarks, and present findings through undergraduate research dissemination channels at Warwick.

Proposed Outputs

• A working Rust-based prototype integrated with a Scientific Python workflow
• Quantitative benchmarks demonstrating performance trade-offs
• Open-source, well-documented research code
• A short research report reflecting critical evaluation of systems-level design choices

The project offers a high-quality undergraduate research experience while contributing insight into the role of modern systems programming languages in accelerating scientific Python software an issue of broad relevance across EPSRC-funded research domains.

Supervisor: Gary Fowmes

Intern: Jennifer Tofts

Department: Engineering

In recent years, polymer-based materials have been heralded as a more sustainable alternative to traditional materials like clay and gravel within groundworks infrastructure by those within its industry. These are known as geosynthetics. They have been shown to reduce the Embodied Carbon of a structure by up to 37% (Xue S et al 2025), consequently reducing its carbon emissions. An aspect by which sustainability is measured is carbon emissions, thus through this lens the geosynthetics industry has labelled itself as “Leading the way to a resilient planet” (Fontana F et al 2023). However, these analyses neglect the long-term implications of utilizing geosynthetics, that of their inevitable degradation and release of harmful microplastics to the environment. This project will aim to experimentally determine the rate at which various geosynthetic materials produce microplastics, more thoroughly examining their overall impact to the environment.

Tests will be conducted on various geosynthetic materials like that of CNC Tensar, who have expressed a desire to collaborate on this project. By testing under laboratory conditions created to replicate in-situ conditions, a high level of control over variables such as the testing period, soil chemistry, loading conditions, exposure to UV rays, and exposure to moisture can be obtained. Furthermore, repeatability can be ensured allowing for comparison between results, while also providing a cheaper method of obtaining results as opposed to testing in the field. For instance, a key construction phase in which microplastics may be produced is the Installation phase, which could be recreated at a desk scale using readily available equipment from the Engineering department. Another area of concern is the interaction of geosynthetics with the coastal environment, utilising engineering’s flume equipment.

Supervisor: Ellen Luckins

Intern: Joshua Thomas-North

Department: Mathematics Insitute

Salts inside the pores of stone or brick can cause damage by a process called salt-weathering. In a conservation context, salt-weathering of historical stone buildings and artefacts must be prevented as far as possible.

To test a stone object for salts, conservationists apply a “poultice” (typically made of several layers of wet filter paper) to the stone surface [1]. The water from the poultice is absorbed into the stone by capillary (surface-tension) forces and salts are dissolved; then as the water dries it is drawn back to the stone surface and into the poultice, bringing the dissolved salts with it. Any salts in the poultice at the end of this process must have come from the stone.

This poulticing method is a straightforward, non-destructive way to detect salts in stone. However, there is currently very little quantitative understanding of the process. Researchers at English Heritage want to understand how to interpret salt measurements from the poultice in terms of the existence, concentration, and location of salts within the stone.

In this project we will model the poulticing process to build predictive capability. We will develop a partial differential equation (PDE) model for the initial absorption of water into the stone, adapting existing models for unsaturated porous-media flow [2]. We will solve this model using numerical and analytical techniques to understand how far water penetrates into the stone. Time permitting, we might couple this water-absorption model with a dissolution model for the salt, and/or investigate the drying stage and the associated salt transport into the poultice.

This is an applied mathematics project, with an interdisciplinary aspect (links to engineering, materials science, and chemistry). It fits centrally within the EPSRC remit.

The student will:

1. Conduct a guided literature review:

a. Mathematical modelling of porous media flows

b. Poulticing studies in the conservation literature, including discussion with conservationists at English Heritage.

2. Construct a mathematical model for the poulticing process, likely consisting of a partial differential equation for the temporal/spatial variation of the water volume-fraction in the stone.

3. Perform a dimensional analysis of the poulticing model to understand the expected force balances and dominant behaviours, and isolate key parameter groupings.

4. Solve the poulticing model using a variety of methods:

a. Numerical solution (scientific computing),

b. Analytical/asymptotic solutions are likely to be possible in certain cases or parameter limits.

5. Interpret results from 3 and 4 to provide new insight into the poulticing process.

6. Communicate results:

a. Write a technical report of the mathematical model, analysis, and results.

b. Create a non-mathematical presentation (eg: slides or poster) to explain and interpret the key findings.

c. Discuss findings and implications with colleagues at English Heritage to maximise impact.

The project outputs are 6.a-c above. The student’s written report might be adapted/extended to form a journal article following the project.

[1] Auras, M. (2008). In Salt weathering on buildings and stone sculptures: proceedings from the international conference, Copenhagen. (pp. 197-217).

[2] Fowler, A. C. (2011). Mathematical geoscience (Vol. 36). London: Springer.

Supervisor: Nicholas Grant

Intern: TBC

Department: School of Engineering

This project seeks to address whether hafnium oxide (HfO2), a common dielectric film used in billions of semiconductor devices, has a detrimental impact on device performance. As such this project will systematically explore the recombination properties of HfO2 on silicon. Previous reports have suggested that defects are created at the interface between the HfO2 film and silicon surface, but recent investigations by the primary supervisor indicate this might not be true, implying defects might also be generated inside the silicon material itself. To ascertain whether the HfO2 layer is negatively impacting the silicon material itself, a charge decoupling method (CDM) will be employed [1]. This novel technique enables us to independently examine changes at the interface and in the silicon material, whereby the HfO2 layer is not removed or damaged in any way.

• Week 1:
  • Literature reviewing, understanding basics of surface passivation and the CDM.
  • Experimental training on atomic layer deposition (ALD), photoconductance decay (PCD), photoluminescence imaging (PL), corona charging (CC) and thermal annealing.
• Week 2:
  • Literature reviewing, understanding basics of surface passivation and the CDM.
  • Testing new experimental skills developed during week 1 (on aluminium oxide (Al2O3) coated samples).
  • Apply the CDM to Al2O3 coated test samples and compare results to those previously obtained by experienced researchers.
• Week 3:
  • Independent data analysis and critical thinking ability after applying the CDM.
  • Once the supervisor is satisfied, the UG student will begin to apply the CDM to HfO2 coated silicon samples.
  • Deposit Al2O3 layers onto HfO2 coated samples via ALD followed by thermal annealing (does hydrogen from the Al2O3 annihilate defects at the HfO2/Si interface and/or in the silicon materials).
• Week 4:
  • Data analysis and critical thinking.
  • Apply the CDM to HfO2 coated silicon samples.
  • Deposit Al2O3 layers onto HfO2 coated samples via ALD followed by thermal annealing.
• Week 5:
  • The same plan as week 4.
• Week 6:
  • The same plan as week 5.
• Week 7:
  • Data analysis and critical thinking.
  • Prepare figures for journal publication.
  • Apply the CDM to HfO2 coated silicon samples.
  • Deposit Al2O3 layers to HfO2 coated samples followed by thermal annealing.
• Week 8:
  • Data analysis and critical thinking.
  • Prepare figures and develop storyline for journal publication.
  • Explore other experimental techniques that could be conducted to aid the journal publication (and that could be conducted by PhD students within the group).

Proposed output: Aside from the UG student acquiring new knowledge and experimental skills, the main output would be to develop research that can be used in a journal publication. While it is unlikely that a journal paper can be written and submitted within an 8-week period, the work to be conducted is novel and thus worthy of publication in an international journal.

Fit to EPSRC remit: The charge decoupling work on HfO₂ fits naturally within the EPSRC remit because it sits squarely in materials science, electronic engineering, and applied physics—all core EPSRC domains.

Supervisor: Feng Hao

Intern: Ahmad Kurdy

Department: Computer Science

This project investigates the application of the Owl augmented password-authenticated key exchange (PAKE) protocol to the problem of secure cryptocurrency wallet authentication.

Cryptocurrency users face a fundamental tension when it comes to storing their money. They are stuck between 2 options: non-custodial (essentially meaning self-owned) private keys give you full control but have 3 major drawbacks: the risk of irretrievable loss of money if you lose your private key, their steep learning curve in using them and that making transactions can be very slow. On the other hand, a user could choose to delegate these responsibilities to a 3rd-party by using a remote wallet which also allows for quick transactions, but the drawbacks here are that this introduces avenues for server-side attacks.

An augmented PAKE allows a user to establish a secure cryptographic key with a server, ensuring the server only stores a mathematical verifier of the correct password instead of the actual password. This makes server-side attacks ineffective because if the server's database is breached, hackers cannot use the stolen verifiers to impersonate the user.

Current remote wallet solutions that rely on PAKE protocols such as SRP-6a, which lacks formal security proofs and elliptic curve support, or OPAQUE, which has implementation challenges and a password-change leakage vulnerability, are that which this project seeks to improve upon. Owl addresses these limitations with provable security assumptions, agility across both multiplicative and elliptic curve group settings, and crucially, zero leakage of password change information to observers.

The project begins with a literature review covering the landscape of cryptocurrency wallets (hot, cold, warm, hardware, etc.) as well as their key properties. This will identify the most appropriate wallet category/categories for Owl integration, since the authentication requirements differ substantially: a hot software wallet accessed via a web interface presents different constraints than a hardware wallet with a USB interface. The literature review will also cover existing PAKE implementations to inform design choices for the Rust implementation.

The core deliverable is a Rust library implementing the Owl protocol, including registration, login, and password update flows as specified in the Owl paper and a paper to support the choices made in delivering this Rust library and the tangible outcomes from using it.

The programming language Rust is chosen for its memory safety guarantees along with its efficient speed close to C’s and for its growing adoption in blockchain and cryptographic software. The library will be implemented against an appropriate cryptographic backend, e.g., curve25519-dalek.

Security benchmarking will simulate server-compromise scenarios to verify the whole point of the PAKE, which is that stolen server data provides no shortcut for an attacker to maliciously gain access to a user’s wallet. Performance testing will measure login latency and bandwidth consumption, comparing results against published figures for SRP-6a and OPAQUE.

The project fits within the EPSRC remit in cybersecurity and trustworthy digital infrastructure, contributing a practical, open-source reference implementation of a formally verified protocol that could directly reduce the risk of large-scale cryptocurrency theft arising from server compromise.

Supervisor: Lukas Eigentler

Intern: Amelie Sims

Department: Mathematics Institute

This project will analyse a new hybrid PDE–ODE model with impulsive stochastic forcing representing trait-structured dynamics. It will use mathematics to elucidate evolutionary dynamics of populations subject to stochastic extreme events. The project will use the evolution of fire resistance in savanna trees as a case study. Fires, fuelled by grasses, facilitate the coexistence of grasses and trees in savannas. Unlike the otherwise superior tree population, grasses can resprout from sub-soil bud banks post fires. This facilitates their long-term persistence. However, trees have evolved fire resistance (e.g., thicker bark) which can affect the delicate coexistence balance.

The student will analyse a novel framework, already developed by the project supervisors, comprising a trait-structured PDE, an ecological ODE and impulsive forcing. Reaction terms comprise ecological competition dynamics between grasses and trees. Evolution of tree fire resistance is described by trait diffusion, akin to phenotypically structured models of cancer cell evolution. Further, the model is subject to impulsive forcing, representing fires, which instantaneously reduce grass and tree biomass. This is modelled as an inhomogeneous Poisson process with an interarrival time dependent on grass biomass. Crucially, the model assumes a trade-off between tree fire resistance (determining how much biomass is lost during fires) and tree growth rate between fires.

The student will analyse this model numerically and analytically. The main aim is to investigate how differences in the functional form of the fire resistance/growth rate trade-off affects the bifurcation structure. For the numerical analysis, the student will implement simulation regimes for impulsive PDE/ODE models and the bifurcation structure will be characterised in terms of first exit times from states. The student will also use pen-and-paper-style analysis of a reduced deterministic model derived from the stochastic framework using averages. This will result in an impulsive model with periodic forcing, whose bifurcation structure will be determined by techniques for periodic or impulsive systems (e.g. Floquet-type methods).

Completion of this project will create preliminary data for the publication of a research article in a leading peer-reviewed journal, such as Journal of Mathematical Biology, potentially after further work by the project supervisors. The main outputs of the project will thus be (i) computational code for the stochastic impulsive model; (ii) bifurcation diagrams for the stochastic and deterministic model; (iii) project report that will act as preliminary data for a publication.

This project sits squarely within EPSRC’s remit in applied mathematics. It develops new theory for nonlinear, multiscale dynamical systems describing trait evolution under impulsive disturbance. While motivated by fire-driven savanna ecosystems, the mathematical framework and analysis methods will remain general and adaptable to other eco-evolutionary systems subject to extreme events.

Timeline:

• Interdisciplinary literature review (weeks 1-2)
• Numerical bifurcation analysis of stochastic model (weeks 2-5)
• Analytical bifurcation analysis of deterministic model (weeks 4-7)
• Report writing (weeks 7-8)

Supervisor: Sharmila Balamurugan

Intern: Thane du Preez

Department: Physics

Aim of the project:

The main goal of this project is to identify any clear and direct signature of quantum vacuum fluctuations of the electromagnetic field. This could be achieved on reaching the following goals.

1. To compare the classical dynamics of coupled, oscillating charged particles with the quantum counterpart:

(a) Classical dynamics: We will consider damping, external drive to sustain motion, and possible anharmonic motion of the particles with nonlinear coupling between them. This could lead to some interesting emergent properties such as synchronisation.

(b) Quantum dynamics: We will study the dynamics of coupled, oscillating charged particles in the presence of quantum vacuum fluctuations of the electromagnetic field. This will help the student understand synchronisation in the quantum regime.

Parts (a) and (b) have been studied in the context of certain specific models such as coupled van der Pol oscillators [for instance, T. E. Lee and H. R. Sadeghpour, Phys. Rev. Lett. 111, 234101 (2013)]. In this paper, they even discuss the effect of quantum fluctuations on synchronisation. Turning the problem around to use the dynamics as a tool to detect vacuum fluctuations should be straightforward. Further, going beyond the van der Pol oscillators to more sophisticated models, could reveal systems more sensitive to the presence of vacuum fluctuations.

(c) We will identify experimentally-accessible physical quantities that carry clear signatures of vacuum fluctuations.

2. To exploit any possible advantage in the context of ease of detection, with increasing the number of particles, as we are studying a collective phenomenon.

Plan of Activities:

Week 1-2: Reading and understanding prior work [T. E. Lee (2013) and references therein]; Developing an in-depth understanding of the concepts of synchronisation in the classical and quantum context; Getting experience with techniques typically used in understanding dynamical systems by reproducing the relevant calculations in [T. E. Lee (2013)]; Identifying the physical quantity that carries a signature specific to the presence of vacuum fluctuations.

Week 3-4: Developing a more general mathematical model of two coupled oscillating charged particles that incorporate external drive, dissipation and nonlinearities in the trapping potential; Studying the classical synchronisation properties.

Week 5-6: Studying the dynamics of the quantum counterpart and the quantum synchronisation properties; Identifying the physical quantity that best captures the presence of vacuum fluctuations; Gauging if there is any advantage in increasing the number of charged particles.

Week 7-8: Writing up the final report and start writing a paper on our results. (Any overflow from Week 5-6 can also be handled here).

Proposed outputs:

1. With the recent experimental realisation of synchronisation in coupled van der Pol oscillators [Yi Li et al., Sci. Adv. 11, eady5649 (2025)], our results would be of topical interest. It could be published in a reputed scientific journal.

2. The student could also present the work in conferences such as TCP 2026 (https://iop.eventsair.com/tcp2026/) and Photon 2026 (https://www.photon.org.uk/).

Supervisor: Brett Kolesnik

Intern: Tommy Harrison

Department: Statistics

Fit to EPSRC remit. This proposal fits within the mathematical sciences, with connections to probability, combinatorics, geometry, algebra, and statistics.

Background. The permutahedron Pi_n is a fundamental object in combinatorial geometry, obtained as the convex hull of the point (1,2,…,n) and its n! permutations.

The permutahedron has connections with tournaments in graph theory. A tournament is an orientation of the complete graph K_n, where each edge of K_n is oriented in one of the two possible directions between its endpoints. We think of the n vertices of K_n as teams, and orient an edge {i,j} from i to j if i wins against j. There are also connections to the statistical theory of paired comparisons, where i winning against j is reinterpreted as one product (or candidate, etc.) being preferred over another.

The score sequence (s_1,…,s_n) lists the total number of wins by the teams. Classical results by Rado (1950) and Landau (1953) show that score sequences are in bijection with lattice points in Pi_n. On the other hand, describing the set of tournaments with a given score sequence is more difficult, and has a long history beginning with Spencer (1974). Since it is difficult to describe these sets directly, it is of interest to sample from them. This was achieved using rapidly mixing random walks in work by McShine (2000), which allows for efficient Markov chain Monte Carlo sampling.

Project aims. We aim to find a continuum analogue of McShine’s result for random tournaments. In a random tournament, the orientation of each edge {i,j} is determined by a probability p(i,j). Such a tournament has a mean score sequence (x_1,…,x_n) listing the expected number of wins by the teams. We tentatively refer to this continuum object as the “Brownian cogs sampler” as it should involve {n \choose 2} many (one for each edge in K_n) copies of reflecting Brownian motion on the interval [0,1]. Brownian motion is the continuum analogue of random walk. These processes should act like “cogs” as they are highly dependent: the movement of one will affect the movement of all others, like in a system of gears. It will be interesting to see how the behavior of this sampler changes as the mean score sequence of the random tournament varies.

Plan of activities. Weeks 1–3 will be spent learning background material on Markov chains, mixing times, random walks, and Brownian motion. In Weeks 3–4 the student will run computer simulations to obtain visualizations and intuition for the problem. By Week 5 we will aim to show that the Brownian cogs sampler exists. In Weeks 6–7 we will analyze its mixing time. During the final Week 8, the student will organize all findings.

Proposed outputs. The student will create a poster and write a final written project, that will lay the foundation for a future article to be submitted for publication.

Supervisor: Allen Hart

Intern: Tsz Dean Chang

Department: Computer Science

Modern large language models (LLMs) are increasingly deployed in high-stakes settings, yet their internal reasoning processes remain mysterious. Mechanistic interpretability is a rapidly developing subfield of AI safety research that aims to reverse-engineer the internal representations and computations of neural networks and render them interpretable to humans; like neuroscience for LLMs. A key recent breakthrough is the sparse autoencoder (SAE), an unsupervised method that decomposes a model's dense internal activations into sparse, human-interpretable features. These features are individual directions in the latent space that correspond to recognisable concepts such as languages, named entities, or syntactic structures. For example the word 'eye' in multiple languages shares the same feature as the eyes in this face O_O.

This project will introduce an undergraduate student to the foundations of mechanistic interpretability through hands-on work with a modern open-weight language model (Gemma 3 1B) and its associated pre-trained SAEs. The project addresses the EPSRC remit in Artificial Intelligence and specifically the priority area of safe, trustworthy, and responsible AI. Understanding internal model representations is foundational to efforts to verify, audit, and align AI systems, and training the next generation of researchers in these methods is itself a contribution to the field's capacity.

The first phase of the project (weeks one and two) will establish the computational environment. The student will set up the TransformerLens and SAELens libraries, load Gemma 3 1B with Gemma Scope 2 SAEs, and reproduce known interpretable features from the literature, gaining fluency with the core workflow of activation extraction, feature identification, and visualisation.

The second phase (weeks three to six) will consist of a focused research investigation led by the student's interests in consultation with the supervisor. One example direction is the geometric structure of number representations. Recent work has shown that some language models arrange numbers along a helix in their latent space, combining linear magnitude information with periodic Fourier-like structure. Causal validation through ablation experiments would test whether identified structures are genuinely used by the model for arithmetic.

The final phase (weeks seven and eight) will focus on writing up findings, preparing visualisations, documenting code as a reproducible open-source repository, and delivering a presentation. The primary outputs are a well-documented codebase, a technical writeup of results, and the student's development as a researcher equipped with skills at the frontier of AI safety.

Supervisor: James Duncan

Intern: Amy Lawrence

Department: Chemistry

The need for novel antibiotics has never been more urgent, as the rise in antibiotic resistance (AMR) is set to lead to an estimated 40 million deaths by 2050 if new treatments are not developed. Microorganisms such as bacteria and fungi are an excellent source of inspiration for new antibiotics, as they produce many bioactive natural products. If we can gain an understanding of how these natural product molecules are biosynthesised, we may be able to bioengineer these systems to make new life-changing medicines. This project aims to investigate how one type of enzyme called an enoyl reductase (ER) domain, is able to carry out an important biochemical transformation, the reduction of an alkene to an alkane, during the biosynthesis of two recently discovered antibiotics, gladiolin and gladiostatin, produced by the bacteria Burkholderia gladioli. Through the chemical synthesis and biochemical assays, this project will uncover whether ER domains can catalyse the reduction of non-natural substrates; this would establish whether it could be used for bioengineering novel antibiotic derivatives, a critical step in addressing the AMR crisis.

This project falls into the EPSRC's ‘chemistry’ remit, specifically within the ‘chemical biology and biological chemistry’ research area, which supports the development and application of ‘novel chemical tools and technologies for the understanding of biology and the synthesis of biological and biologically active molecules’. The project will heavily feature organic synthesis, analytical chemistry (mass spectrometry (MS), nuclear magnetic resonance (NMR)) and enzymology, which are all core methodologies in biological chemistry. This aligns with the EPSRC's stated priority to ‘encourage the development of the next generation of platform technologies to enable the pursuit of new academic challenges and exploration of previously inaccessible chemical and biological space’.

The project will be conducted over 8 weeks in the Chemical Biology Research Facility (CBRF) at the University of Warwick, within the Challis/Alkhalaf research group.

Weeks 1-5: Substrate synthesis: The student will design and carry out multi-step synthesis of a panel of 2-enoyl pantetheine thioester substrates representing structural analogues of the natural ER substrate. This will involve dry and Schlenk line techniques, column chromatography for purification, and structural characterisation of products by NMR spectroscopy and MS.

Weeks 6-8: Biochemical enzyme assays: The synthesised substrates will be tested in a functional in vitro enzyme assay using already produced and purified recombinant ER domain proteins. High-resolution MS will be used to detect and quantify enzymatic reduction of each substrate, thereby establishing the substrate scope of the enzyme.

Novel insight into the substrate tolerance of the gladiostatin and/or the gladiolin ER domains will be established; this will feed directly into ongoing research within the group on bioengineering of these biosynthetic pathways. The student will produce a written project report and deliver a short presentation to the research group. They will also be supported to contribute to any resulting publications or patent applications. Longer-term, these findings will inform the rational bioengineering of modified polyketide assembly lines as a route to novel antibiotic candidates, contributing to the UK's strategy for combating AMR.