This page contains the abstracts of all accepted submissions for the STEP-UP RSLondon Conference 2026. You can browse the abstracts here or link to a specific abstract by clicking a talk title in the conference schedule.

Where available, links to conference presentations are also provided.

Poster abstracts are provided towards the bottom of the page.

Title: What Remains Without Instructions: The Transfer Service and the Preservation of Obsolete Digital Media
Authors: Leontien Talboom, Cambridge Digital Humanities
Type: Keynote
Time/Track: 10:10-10:55, General track
Abstract:
Digital preservation is often framed as a technical problem of maintaining files and formats, but in practice it also depends on preserving the knowledge needed to access and interpret obsolete media. This talk explores the work of the Cambridge University Libraries and Archives Transfer Service in recovering data from legacy carriers such as floppy disks, highlighting the practical and often undocumented expertise involved in making old digital objects readable again. It argues that what remains without instructions is not just data at risk of loss, but also the gradual disappearance of the skills, software, and workflows required to recover it.



Title: How many double-deckers does it take to ruin your projects?
Authors: Rosie Wood, Mosè Giordano, Mahmoud Abdelrazek, Sam Cunliffe, David Pérez-Suárez and Markus Hauru
Type: Regular talk
Time/Track: 11:45-12:00, General Track
Abstract:
Are you interested to know about the impact of your open-source projects in your organisation? Are you worried that projects are being abandoned and accumulating dust (issues) that nobody is cleaning? Are you sure your open source projects are actually open source licensed? How many people need to disappear before your projects become ghosts? If you are comfortable with your answers to all of these questions then you don’t need to attend our talk. Otherwise, we are here to help you to avoid future disasters.

We’ve tried many tools that claim to provide answers to the above so you don’t need to suffer like we did. In this talk we will give an overview of them, their pros and cons and when you would (or not!) want to use them.



Title: Collaborative Computational Project for Volume Electron Microscopy (CCP-volumeEM): Connecting People, Tools, and Data Across the Volume EM Community
Authors: Martin Jones
Type: Regular talk
Time/Track: 12.00-12:15, General Track
Abstract:
We introduce CCP-volumeEM, a newly formed Collaborative Computational Project bringing together researchers, microscopists, and software developers across the diverse landscape of volume EM imaging in biology. As acquisition technologies advance, the analytical challenges facing volume EM necessitate handling terabyte-scale datasets, with computational tasks including 3D reconstruction, registration, and segmentation. CCP-volumeEM addresses these challenges through three strands: community building, research software and data management, and training. We are supporting and developing open, sustainable software solutions, promoting FAIR data principles, and promoting best practice in the adoption of new technologies. Through hackathons, training sessions, and collaborative events, we aim to accelerate progress and widen access to cutting-edge methods across the biological EM research community, while providing persistent resources via our website (https://www.ccp-volumeem.ac.uk/) and other channels.



Title: Bridging the Gap: Why Research Community Managers are the Human Essential in an AI-Driven Landscape
Authors: Cassandra Gould van Praag, Emma Karoune, Malvika Sharan and Danny Garside
Type: Regular talk
Time/Track: 12:15-12:30, General Track
Abstract:
Overview: This talk highlights the role of community management and strengthens the argument that Research Community Managers (RCMs) are an increasingly important component of the dRTP ecosystem. While LLMs and other AI tools are increasingly used to support generation of code, documentation and data processing, they cannot replicate the RCM’s core functions of establishing connections, building trust, helping navigate institutional politics, fostering cross-disciplinary collaboration, and building the socio-technical infrastructure necessary for solving complex problems. Indeed recent labor market data from the World Economic Forum (2025) and OECD (2024) confirm a ‘human-centric’ pivot which places an increasing premium on ‘bridging’ roles that require coordination, interpersonal judgement and emotional intelligence. This evidence suggests that the RCMs, either in formal or informal roles, are a human necessity in an AI-enhanced research lifecycle, providing the strategic cohesion AI lacks

Interest to Attendees: Workshop attendees are in a time of significant shifts in employment trends. This talk provides a data-backed argument for why RCMs are a strategic necessity for institutions to maximise their technical investments. This will be valuable to junior dRTPs who are developing career strategies, and senior leaders looking to build maximum (and future proofed) impact in their projects.

Impact and Discussion: This presentation will develop attendees’ understanding of how to “AI-proof” their career structures by focusing on non-automatable skills. It aims to spark a critical discussion on institutional policy: how do we move beyond “ad-hoc” community management to formal, senior-grade pathways? By the end of the session, attendees will be equipped with a framework to advocate for RCM roles as the essential link between shifting technological tools and the long-term sustainability of the research lifecycle.



Title: Preliminary results from mapping and evaluating dRTP’s contributions to computing skills and pedagogies in HE
Authors: Timothy Monteath and Carlos Cámara-Menoyo
Type: Lightening talk
Time/Track: 12:30-12:35, General Track
Abstract:
Our research – funded by the UKRI+ DisCoRSE network seeks to map and analyse the provision of digital research competencies (DRC) which are led by or involve dRTP’s across higher education institutions in the UK. The teaching of computational methods has spread across disciplines, spawning numerous pedagogical approaches and a wide variety of methods, software, products, and environments. However, the role of dRTP’s in the proliferation teaching DRC remains unclear. dRTP’s responsibilities are primarily focused on research and infrastructure and expectations around involvement in training and teaching vary widely across the sector. Furthermore, choices as to which approaches and delivery modes to adopt in teaching and training are often driven by trends, licensing constraints or trainer’s familiarity rather than pedagogical criteria or context suitability.

Our research therefore seeks to investigate what training and teaching is being delivered by dRTP’s? What skills and competencies are being taught? What pedagogical approaches are being taken? How does this differ across audiences, infrastructures, and institutions? We will use this lighting talk (and the poster we are also proposing) to report on our preliminary findings from a survey we have been running and audit of curricula and training involving dRTP’s listed on university websites and to further develop conversations with the dRTP community over their involvement in education and training.



Title: The DIRECT Framework
Authors: Adrian D’Alessandro
Type: Lightening talk
Time/Track: 12:35-12:40, General Track
Abstract:
In this lightning talk I plan to introduce the DIRECT Framework and how it can be useful for dRTPs, along with my SSI Fellowship and a possible announcement of a RSLondon hackathon event to contribute to the DIRECT Web App.

The DIRECT framework is a framework of skills and competencies for dRTPs aimed at helping with career development. A central component of delivering this is the web app which so far has been developed by a collection of volunteers from the RSE community. The Web app provides users a way to generate a visualisation of their own skills profiles by determining what skill level they have in all of the different skills. In future, users will be able to compare these profiles to reference profiles of dRTP jobs at different career levels to inform their professional development.

The goal of my SSI Fellowship is to increase the number of open-source contributors to the DIRECT web app. This will be achieved by lowering the barrier for entry to contributing through: simplifying to code structure, improving the developer documentation, and inviting people to contribute at hackathon events, where new contributors will get hands-on help.



Title: Portability of Using C++20 in R Packages
Authors: Sherman Lo
Type: Regular talk
Time/Track: 14:00-14:15, Software/RSE Track
Abstract:
The programming language C++ is still widely used for high-performance computing. There are numerous efforts to modernise the language with new features to improve the C++ experience, such as memory-safety features and making code less verbose. However, implementing these new features in your code too early may cause problems on systems which may not have caught up with the new C++ standards.

As a case study, we will look at our R package, which does clustering of polytomous variables. We’ve used hybrid programming so that our package has a familiar R interface while running high-performance multi-threaded C++ code for the underlying calculations. We have successfully submitted the package to CRAN, but there were a few friction points regarding portability.

During maintenance, we wanted to use some of the new C++20 features, such as std::span and std::jthread. However, when CRAN checked our package, we found it failed on their MacOS systems - in particular, those with Clang versions 15 or lower. It failed because some C++20 features were either partially implemented or not at all, causing failures on the CRAN checks. To fix this, we’ve refactored our code by removing the offending C++20 features. With version control, these can be reverted when newer MacOS systems fully adopt the new C++ features.

In terms of future work, we plan to use newer C++20 and C++23 features to improve the code base. However, we also have to balance portability; to roll out these new practices once the majority of systems have implemented these new features. Being too early an adoptor could cause your software to be out of reach for the majority of your users - an important message for those programming in C++ or users of an R package which uses C++.



Title: The Python Array API Standard
Authors: Patrick J. Roddy
Type: Regular talk
Time/Track: 14:15-14:30, Software/RSE Track
Abstract:
The Python Array API standard is a way to standardize how different array libraries should work. Many developers are familiar with the nuances of NumPy—for example, how one might achieve the maximum value of an array. However, with the rise of deep learning, we now have many new array types including TensorFlow, PyTorch, and JAX. Some of these libraries follow the precedent set by NumPy, while others choose to implement their interfaces differently.

For library maintainers and newer developers, this inconsistency can lead to confusion. Furthermore, if a library intends to support multiple backends, they often have to handle many names for similar functions. This is where the Array API standard improves the Python ecosystem. By providing a consistent API for all libraries to adopt, it allows downstream tools to work with any array type. All internal computations are performed directly by the incoming library, meaning there is no extra cost involved in swapping between types.

The standard does not strictly follow the NumPy API, as NumPy was not originally designed with non-CPU devices, graph-based libraries, or JIT compilers in mind. At UCL ARC, we have been working on porting a NumPy-based codebase to the Array API. Because the ecosystem is still fairly new, we are frequently adapting to changes within the standard. In this talk, we will share our experience with this transition and the lessons learned along the way.



Title: What RSEs should know about supply chain security
Authors: Mark Woodbridge
Type: Regular talk
Time/Track: 14:30-14:45, Software/RSE Track
Abstract:
Recent high-profile attacks on open-source projects - ranging from malicious dependency injection to compromised build pipelines - have highlighted the importance of supply chain security. While these breaches have been relatively well-publicised, they can easily be dismissed as esoteric or irrelevant to the dRTP community.

This talk breaks down exactly how these attacks are executed and explores their harmful consequences. It examines the relevance of supply chain security to research software, focusing on the intersection of information security, research integrity and data provenance. Using real-world examples from the PharosAI project as a case study, the session provides actionable advice for RSEs to secure their workflows. It is applicable to any project utilising third-party frameworks, libraries or tools, and it will be especially relevant for those working on collaborative projects.



Title: Beyond language silos: purely functional package management with Guix
Authors: Arun Isaac
Type: Regular talk
Time/Track: 14:45-15:00, Software/RSE Track
Abstract:
Modern software is complex. There is an endless multitude of dependencies spanning several programming languages, and we end up with a panoply of package managers that try to tie them all together in some comprehensible way.

It has become widely accepted practice for every programming language to have its own package manager and related tooling. But these package managers struggle to straddle programming language boundaries and often end up shipping pre-built binaries undermining the whole point of package management. As a result, reproducibility suffers and it increasingly seems that we have little choice but to resort to shipping binaries everywhere. The dream of the distro as the single authoritative source of packages seems but a quirk of the past. But there is still hope.

In this talk, I will introduce Guix, a purely functional package manager. This simple principle leads to many interesting consequences. Guix packages are a pure function of their source and build recipe, providing strong reproducibility guarantees. Binary substitutes are simply a memoization of this function. Local package definitions are just another function—and thus just as first-class as packages provided by upstream repositories.

Like functional programming languages managing memory, Guix installs packages to an immutable content-addressed store. This means that many disparate features such as rootless installs, language-agnostic virtual environments and rollbacks all fall out naturally rather than being bolted on as an afterthought. Software can be simple, comprehensible and fit in one’s brain again.



Title: A DevOps-Inspired Open-Source Platform for Harmonisation, Quality Control, and Analysis in Multi-Site Ultra-Low-Field Neuroimaging
Authors: Hajer Karoui
Type: Lightening talk
Time/Track: 15:00-15:05, Software/RSE Track
Abstract:
Multi-site neuroimaging studies face persistent challenges in data harmonisation, quality control, and accessibility of processing pipelines, challenges that are amplified in global networks operating across diverse technical environments. Within the Ultra-Low Neuroimaging in the Young (UNITY) network, over 40 portable 64mT Hyperfine Swoop MRI systems have been deployed across high- and low/middle-income country (HIC/LMIC) settings, generating large volumes of ultra-low-field (ULF) paediatric brain data. While Flywheel, our medical imaging data management and analysis partner platform, provides centralised storage and containerised processing, downstream stages including data retrieval, cleaning, and quality control remained technically demanding and inaccessible to non-developer collaborators.

To address this, we developed an open-source, locally deployable Streamlit web application extending the UNITY Flywheel ecosystem. The platform was built following software engineering best practices: modular architecture, version control via GitHub, continuous integration and deployment, and iterative design guided by feedback from data managers and researchers across LMIC and HIC sites. This collaborative, user-driven development process ensured the platform addressed real operational needs rather than assumed ones.

The application provides authenticated, browser-based access to multi-project data without requiring command-line expertise. Implemented modules include harmonised derivative retrieval across projects, statistical outlier detection and data cleaning, interactive brain segmentation QC visualisation, and batch execution of analysis gears. All outputs log MRI software version, gear version, and Flywheel analysis ID, ensuring full provenance traceability.

Preliminary deployment across UNITY sites demonstrated substantial gains in workflow efficiency. Tasks previously requiring manual SDK scripting are now executable through the interface in significantly less time, while maintaining reproducibility. By lowering the technical barrier to data management and quality control, the platform democratises participation in neuroimaging research across sites with varying RSE capacity — directly demonstrating the value of applying software engineering best practices to real-world, equity-focused research infrastructure.



Title: How to stop doing web scraping all wrong
Authors: Giles Greenway
Type: Lightening talk
Time/Track: 15:05-15:10, Software/RSE Track
Abstract:
Web scraping is a common practice in both research and data engineering, after all, data has to come from somewhere. Unfortunately, much of the advice on how to do it is of rather poor quality, especially in Python for novice developers. Low-level tools for making HTTP requests and parsing HTML are often suggested. These are superficially easy to use, but it is also all too easy to write badly structured code. The Scrapy framework is often presented as having a steep learning-curve. It has nothing of the sort, rather it provides a safety rail that stops novices getting into trouble, and features they don’t know they need.

Furthermore, scraping need not have anything to do with parsing HTML. A simple, instructive example using historical Parliamentary petition data will show Scrapy succeeding where Requests and BeautifulSoup fail. Researchers who have followed the pervasive bad advice have had very bad and frustrating experiences. Their first taste of research software development should not be like this.



Title: CODECHECK: a system for the independent reproduction of results reported in scientific publications
Authors: Stephen Eglen
Type: Regular talk
Time/Track: 15:45-16:00, Software/RSE Track
Abstract:
The traditional scientific paper falls short of effectively communicating computational research. To help improve this situation, we propose a system by which the computational workflows underlying research articles are checked. The CODECHECK system uses open infrastructure and tools and can be integrated into review and publication processes in multiple ways. We describe these integrations along multiple dimensions (importance, who, openness, when). In collaboration with academic publishers and conferences, we demonstrate CODECHECK with 25 reproductions of diverse scientific publications. These CODECHECKs show that asking for reproducible workflows during a collaborative review can effectively improve executability. While CODECHECK has clear limitations, it may represent a building block in Open Science and publishing ecosystems for improving the reproducibility, appreciation, and, potentially, the quality of non-textual research artefacts. The CODECHECK website can be accessed here: https://codecheck.org.uk/.

Joint work with Daniel Nuest, TU Dresden



Title: Research Software Sharing, Publication, & Distribution Checklists
Authors: Richard Acton
Type: Regular talk
Time/Track: 16:00-16:15, Software/RSE Track
Abstract:
These research software sharing, publication, and distribution checklists were inspired by similar checklists produced by the imaging community and are intended to address a niche not addressed by existing similar resources such as guidance for creating research software management plans. They provide a tiered approach to completing the checklists with each item able to be completed at a different level:Bronze,Silver,Gold, and Platinum where bronze is highly attainable and platinum is going well above and beyond. With this approach I’m aiming to somewhat gamify the process as well as to provide aspirational goals and not just to set a minimum floor for compliance. The checklists are tailored based on a simple taxonomy of research software output types: ‘records of specific analyses’, ‘web based services’, ‘software packages’, and ‘pipelines or workflows’. Each of these types of software output have overlapping but slightly different considerations for how best they can be shared, published, and distributed so each has its own checklist.

Each checklist is comprised of items which address these 11 common themes: Source control, © Licencing, Documentation, Making Citable, ✅ Testing, Automation, Peer review / Code Review, Distribution, Environment Management / Portability, Energy Efficiency, and ⚖ Governance, Conduct, and Continuity. The checklists are provided as simple markdown files making the checklists easy to include in a project repository like standard licenses and codes of conduct. Each theme includes its checkbox items and an expandable section which expands with advice, links, and resources on how to complete these checkboxes. There are repo badges to display the type of checklist, overall score, medal, and self vs. third party assessment. See rsspdc.org for more details.



Title: Automate the Boring Stuff with AI
Authors: Rebecca Walker and Chris Cheshire
Type: Regular talk
Time/Track: 16:15-16:30, Software/RSE Track
Abstract:
Delivering high-quality bioinformatics software requires coordination across scoping, development, review, and documentation. These processes are often fragmented, manual, and difficult to standardise. We present an end-to-end AI-assisted delivery system that integrates into each stage of the software development lifecycle, from initial project proposal through to automated documentation.

The system is structured around three interlocking layers. First, a scoping agent gathers context from internal, existing codebases and the internet, resolves ambiguities through dialogue with the requester, and produces a structured design proposal with a ready-to-assign set of tasks. Second, during development, specialised agent swarms equipped with domain-specific skills assist with planning, implementation, and code review. These handle syntax and boilerplate so that developers can focus on architecture and design decisions. Finally, after code merge, an automated documentation pipeline retrieves technical content from all relevant repositories, synthesises project summaries, and submits a structured draft for human review. Underpinning all three layers is a principle of composability: the agents and skills built upon existing AI tools are reusable and shareable across projects and teams in a modular, user-configurable way.

This talk will interest attendees who want to move beyond ad hoc AI usage towards structured, reproducible systems that integrate directly into development workflows. No proprietary infrastructure is required; the approach is built on widely accessible tools and transferable design patterns, making it immediately applicable and deployable in various research computing settings.

Attendees will leave with practical insight into designing agentic systems: how to decompose complex workflows into composable agents; how to manage orchestration and context; and how to integrate human oversight at the right points. The talk will also open discussion on a key question for the field: when should AI agents be lightweight and task-specific, and when do they warrant investment as robust, reusable systems?

Data abstracts



Title: Who Does the Data Work? Roles, Recognition, and Career Pathways in Biomedical Data Science
Authors: Denise Bianco, Daria Sokolova, Giulia Tomba, Kim Gurwitz, Vera Matser, Catherine Brooksbank and Emma Karoune
Type: Regular talk
Time/Track: 14:00-14:20, Data Track
Abstract:
The rapid evolution of biomedical research increasingly depends on data-intensive methodologies, creating a growing demand for professionals with robust data science expertise. In the UK, the DSIT National Data Strategy (2023) identified challenges arising from inconsistent definitions of data skills and fragmented descriptors for data science roles. Similarly, the Medical Research Council’s 2022 review of biomedical data science highlighted the need to attract and retain skilled professionals, calling for clearer career pathways and the adoption of team science approaches that enable individuals to be appropriately supported and rewarded. However, career recognition and progression in biomedical data science remain inconsistent, often hindered by vague role definitions and limited acknowledgement of collaborative contributions.

The Advancing Biomedical Data Science Careers (ABDC) project, funded by the Medical Research Council and jointly led by The Alan Turing Institute and EMBL-EBI, addresses these systemic challenges by mapping the diverse skills, roles, and collaborative practices that underpin biomedical data science, while examining recognition and reward of data professionals in team science.

A better understanding of roles and career pathways is essential for strengthening education, recruitment, training, and retention, as well as supporting effective collaboration. In this talk, we will present findings from our ongoing qualitative study investigating interdisciplinary biomedical data science practices in the UK. Through this study, we examine how organisations of different types and scales structure data science teams and collaborative workflows, and how these structures support or constrain successful interdisciplinary practice. Drawing on a diverse cross-section of collaborators and institutional contexts, the study explores challenges, best practices, and future needs for initiating and sustaining effective team science in biomedical research, with a strong focus on career development and progression pathways for data professionals. While the study is ongoing, we will present our methodology and sample, discuss key emerging themes, and outline next steps.



Title: FAIRness Starts at the Point of Creation: How to Implement Electronic Lab Notebooks to Support FAIR Practices
Authors: Samantha Pearman-Kanza
Type: Regular talk
Time/Track: 14:20-14:40, Data Track
Abstract:
FAIR data is often thought about at the end of a project and treated as a box ticking exercise that can be achieved through checklists, automated assessment tools, or the adoption of the right data management tools. However, data cannot be retrospectively made FAIR, and no amount of post processing can compensate for data that was poorly planned or inconsistently captured. This talk explores how FAIRness needs to be built in from the point of creation. It argues that good data stewardship and appropriate research software support are central to this process, rather than relying on tools alone. After all, digitising poor paper based practices does not magically improve them, it just makes them digital.

Drawing on practical experience, it highlights how early decisions about data capture, structure, and workflow shape the long term usability and reusability of data. Using Electronic Lab Notebooks as a practical example, the talk examines how ELNs can be implemented and configured in ways that support and encourage FAIR principles, rather than simply digitising existing practices. It emphasises the importance of thoughtful implementation, user adoption and engagement, and above all continued training and support to enable genuinely FAIRer data.



Title: Energy Efficient Climate Science Workflows with PyActiveStorage
Authors: Varsiha Sothilingam, Bryan N. Lawrence, David Hassell, Valeriu Predoi and Max Norton
Type: Regular talk
Time/Track: 14:40-15:00, Data Track
Abstract:
As climate modelling enters the exascale era, the volume of data produced is outstripping the capacity of traditional networks to transport it. The conventional approach, pulling petabytes of data from remote storage to a local storage, is no longer viable due to high latency and the unsustainable energy costs associated with massive data movement. Data proximate computing, using active storage (also known as computational storage) offers a vital alternative for performing critical parts of the data analysis workflow. Even when true active storage is not possible, many of the advantages of active storage can be achieved with relatively little effort by the storage system providers.

To harness local resources, the PyActiveStorage architecture pushes compute tasks directly down to the storage system or a suitable adjacent system. With an MPI-style reduction interface, complex data operations can be carried out in situ or near situ. The use of a prescribed definition and implementation of a reduction interface for the data, also means security and load level issues can be addressed. As a consequence, smaller data outputs can be returned to the user. By executing these operations at or near the storage layer, the system bypasses the traditional bottlenecks of large-scale data movement.

PyActiveStorage provides a near-production quality implementation of active storage technology, specifically engineered for deployment in high-performance research environments. The software architecture follows a modular provider pattern, allowing the same API to interact with diverse storage backends. Currently deployed on the JASMIN high-performance object store and optimised for NetCDF4 and HDF5 datasets, the library provides a seamless user-facing application for performing data reductions. We will demonstrate how PyActiveStorage simplifies the transition to near-data processing on JASMIN, ensuring that as research datasets reach the exascale, analysis capabilities and sustainability keep pace.



Title: Exploring and Evidencing the need for Data Stewardship - Exploring the outcomes of CaSDaRs First Funding Call
Authors: Louise Saul and Samantha Pearman-Kanza
Type: Regular talk
Time/Track: 15:00-15:20, Data Track
Abstract:
The Careers and Skills for Data Driven Research Network+ (CaSDaR – www.casdar.ac.uk) is a UKRI funded Network Plus geared towards creating a network of data stewards and supporting the careers and skills development of Data Stewards, individuals who oversee the correct collection, curation, structure, and sharing of organisational data within the correct governance frameworks.

A key deliverable of CaSDaR is to highlight the essential skills needed by Data Stewards and to promote the organisational culture change required for these skills to be formally recognised. To support this, CaSDaR has allocated flexible funding to demonstrate the value of embedding Data Stewardship practices within processes, projects, collections, and organisations through funding two different project types. The first were £5k projects lasting for 8-10 weeks, geared towards upskilling prospective trainee data stewards. These projects were intended to give individuals experience in Data Stewardship, supporting the establishment of the role as a viable career option. The second were £70k Projects lasting 6-9 Months; These projects were geared towards upskilling established individuals interested in enhancing their career progress in roles aligned with Data Stewardship.

CaSDaR completed the first call to distribute these flexible funds in in March 2026. In this talk, we will describe how the funding call enabled the CaSDaR team to identify the skills and needs of individuals working in Data Stewardship and of the organisations that employ them. We will also discuss the areas of Stewardship that applicants believe require further development, how organisations currently support Data Stewards, and the expected outcomes of the funded projects. Finally, we will outline how these insights will inform CaSDaR’s ongoing strategy and the delivery of their objectives.

Infrastructure abstracts



Title: Improving the visibility of computing infrastructure with a mini HPC cluster
Authors: Eirini Zormpa, Jeremy Cohen, Weronika Filinger, Neil Chue Hong, Laura Moran, Charaka Palansuriya, Anna Roubíčková and Martin Robinson
Type: Regular talk
Time/Track: 14:00-14:20, Infrastructure Track
Abstract:
Academic research is increasingly reliant upon Digital Research Infrastructure (DRI). DRI encompasses not only data, software and computing facilities, but also skilled digital Research Technical Professionals (dRTPs) who use this infrastructure to support research. In this talk, we will focus on the computing facilities and skills aspects of DRI, specifically how to address the current lack of dRTPs with an HPC focus.

We attribute this partly to the limited visibility of HPC hardware. Where HPC facilities used to be housed locally and often provided graduate students with the opportunity to see, interact with and even manage HPC facilities, these are now based remotely and managed centrally. There are good reasons for this shift, but we believe that it has inadvertently removed opportunities for people to learn about HPC and become skilled in the area.

One way to address the issue is to make this infrastructure more visible. As part of the DRIFT project, we have built a mini-HPC cluster that can be used for training. Crucially, the cluster is portable; it is intended to be in the rooms where the training or demos are taking place to demystify what computing infrastructure looks like and, hopefully, inspire attendees to learn more about it.

At the heart of this portable cluster are 36 Raspberry Pi 5 nodes, of which 32 act as compute nodes with the remaining 4 providing an interactive login node and a distributed storage system. The cluster has been designed to showcase features found on a full-size rack-mounted computing cluster, including an enterprise network switch and power distribution unit, with the hardware mounted in a 19-inch rack case. In this talk we will present the cluster design in more detail as well as training materials we are developing to introduce HPC to people in a range of roles.



Title: ACIT Hub: Co-designing training solutions to support robust, reproducible accelerated compute infrastructure
Authors: Anna Swan and Helen Cooper
Type: Regular talk
Time/Track: 14:20-14:40, Infrastructure Track
Abstract:
Research Infrastructure Engineers (RIEs) are critical to realising the value of UK investment in large scale computing systems: they design, operate and support the platforms that enable AI, data‑intensive research, and large‑scale simulation. Yet, unlike Research Software Engineers, research technology professionals working on accelerated compute lack coordinated, national support for skills development and recognition. The Accelerated Compute Infrastructure Training Hub (ACIT Hub) is working to address this gap.

Co‑design is central to ACIT Hub’s approach. Over three co‑design days we brought together RIEs and related roles to discuss and prioritise both training topics and delivery formats. We present the findings from these community scoping sessions and discuss the gaps in the dRTP landscape for RIE training. The ACIT Hub has already started to fill the gaps, with hackathons and webinars. In our talk, we will showcase our most recent hackathon on performance monitoring and outline what’s to come. We will introduce our training plan and contextualise it within the DIRECT skills framework.

We are keen to share the co-design model, and training focuses of ACIT Hub and continue to engage the community to help shape our training plan. Additional co‑design days are planned for later in 2026 to broaden community involvement and ensure training remains responsive to evolving needs and technologies.



Title: YATRET: Yet Another Trusted Research Environment Talk
Authors: Matt Penn
Type: Invited talk
Time/Track: 14:40-15:00, Infrastructure Track
Abstract:
For the past 5 years the KCL e-Research department have been developing and operating an in-house ISO 27001 certified TRE. This talk aims to provide a broad overview of all the exciting (for us at least) technical capabilities we have deployed to date. We will look toward the challenges and opportunities on the path forward and consider whether some of these architecture patterns have broader applicability in a post-Mythos era (where era ~= 3 - 6 months).



Title: Cloud-Native Without the Cloud: Infrastructure Enabling Researcher Workflows
Authors: Piper Fowler-Wright
Type: Invited talk
Time/Track: 15:00-15:15, Infrastructure Track
Abstract:
Cloud-native technologies allow infrastructure engineers to deploy resilient, automated, and scalable Digital Research Infrastructure, free from vendor lock-in and exorbitant cloud costs. The principles of these technologies also facilitate research-specific workflow delivery. Yet despite increasing adoption in industry, cloud-native tools remain underutilised in many research institutions. The Cloud-Native Special Institute Group addresses this directly, providing training and resources to improve the discoverability of modern infrastructure tools for dRTPs.

We advocate a model of enabling without prescribing: give operators open infrastructure choices and researchers the utility to run their own workflows, but don’t mandate either. dRTPs sit in the middle, bridging infrastructure and researcher workflows.

Two examples from the Rosalind Franklin Institute illustrate this in practice. Our self-managed Kubernetes infrastructure, administered declaratively via GitLab and ArgoCD, gives operators ownership of their deployments. Globus Flows are integral to our instrument data pipelines, moving ~TB daily across multiple storage backends; role-based access controls allow flows to be run automatically or triggered by researchers without requiring direct permissions on the underlying storage. These patterns are transferable: cloud-native is less about where applications are deployed, and more about how. The Cloud-Native SIG exists to help dRTPs make use of it.

Poster abstracts

Title: Adoption of EPIC EHR Data into the OMOP Schema
Authors: Alan Nardo
Type: Poster
Abstract:
Cambridge University Hospitals (CUH) we are interested in adopting the Observational medical Outcomes Partnership (OMOP) schema, an international standard data format for hospital records, to create a means to deliver standardised medical records data to partners for the purpose of conducting clinical research. Our over-arching objective is to create a data repository for this sensitive clinical data, which will be provided to researchers via the NHS Digital East of England Secure Data Environment (a Trusted Research Environment).

This will be done by taking existing electronic health record (EHR) data, stored in Epic databases, and transforming them into the OMOP schema. Being the first NHS trusts to adopt the Epic EHR system, CUH has over 10 years of records, containing tens of millions of rows of patient records. In this process, we have been able to transform the key source tables into OMOP-compliant outputs. Our current pipeline combines R with data held in SQL Server databases, using on-prem NHS servers, but we are currently planning a transition to a Cloud-based research computing infrastructure using Microsoft Fabric.

Our work has involved building the technical infrastructure for this Research Data project within the constraints of NHS systems. We will discuss the technical challenges we have addressed, which will be shared by many digital Research Technical Professionals working with secure NHS data.

Title: Preliminary results from mapping and evaluating dRTP’s contributions to computing skill and pedagogies in HE
Authors: Carlos Cámara-Menoyo and Timothy Monteath
Type: Poster
Abstract:
Our research – funded by the UKRI+ DisCoRSE network – seeks to map and analyse the provision of digital research competencies (DRC) which are led by or involve dRTP’s across highereducation institutions in the UK. The teaching of computational methods has spread across disciplines, spawning numerous pedagogical approaches and a wide variety of methods, software, products, and environments. However, the role of dRTP’s in the proliferation teaching DRC remains unclear. dRTP’s responsibilities are primarily focused on research and infrastructure and expectations around involvement in training and teaching vary widely across the sector. Furthermore, choices as to which approaches and delivery modes to adopt in teaching and training are often driven by trends, licensing constraints or the trainer’s familiarity rather than pedagogical criteria or context suitability.

Our research therefore seeks to investigate what training and teaching is being delivered by dRTP’s? What skills and competencies are being taught? What pedagogical approaches are being taken? How does this differ across audiences, infrastructures, and institutions? We will use our poster to; report preliminary findings, encourage attendees to take part, and further develop the conversations we have been having with dRTP’s about their involvement in training and teaching DRC.

Title: Green Alma: Bringing Carbon Footprint Awareness to HPC Workflows at the Institute of Cancer Research
Authors: Stacy Shcherbakova
Type: Poster
Abstract:
As computational research grows in scale, the environmental cost of high performance computing (HPC) is becoming harder to ignore, yet most researchers have little visibility into the carbon footprint of their workflows. Green Alma is an internal tool developed at the Institute of Cancer Research (ICR) that adapts the GreenAlgorithms4HPC framework for our local HPC environment. It collects job metadata from the scheduler, computes per job and aggregate carbon estimates based on hardware usage, runtime, and regional grid carbon intensity, and surfaces these to researchers through an accessible interface without requiring any changes to how jobs are submitted or run.

Attendees will gain practical insight into adapting an open source carbon estimation framework to a specific institutional HPC environment: the technical decisions involved and the design choices made to keep the tool accessible and meaningful to researchers. The poster will also open discussion on how RSEs can play a central role in embedding sustainability into research infrastructure as a first class concern alongside performance and reproducibility.

Attending the Green SIG at RSECon last year made clear how far ahead ICR is compared to many other institutions, and this poster is an opportunity to share that experience. Green Alma is a practical, researcher facing implementation rather than a theoretical framework. The questions it raises around awareness, behaviour change, and institutional carbon reporting are relevant whether attendees work in HPC, research infrastructure, or software development. We hope it will spark conversation about shared approaches and the role RSEs can play in driving environmental accountability in research.

Title: Demystifying HPC: Empowering Life Scientists in HPC use for BioImage Analysis
Authors: Todd Fallesen
Type: Poster
Abstract:
High Performance Computing (HPC) is fast becoming an important tool for wet-lab scientists, particularly those working in BioImage analysis. Microscopy data analysis has traditionally been performed on personal computers, but with the rise of complex datasets, where terabyte sized mutli-dimensional image sets are increasingly common, data analysis must move to high performance computing.

HPC training has commonly designed for scientists with previous computational experience, who are comfortable interacting with computers at a low-level way, rather than those who have limited scientific computing experience, such as many wet-lab scientists.

We held a one-day workshop at the Francis Crick Institute where we brought together wet-lab scientists, bioimage analysts, bioinformaticians and HPC engineers. The wet-lab scientists presented their motivations for using HPC in their research, where stressed that HPC was seen as more of a tool to finish their biological analysis, rather than something they wanted to learn in depth. Together with the bioimage analysts, bioinformaticians and HPC engineers, the group discussed what skills in computing and HPC wet-lab biologists would need to do their own analysis.

Through this workshop, we realised that any curriculum to teach HPC would have to include some basic computing as well, as many learners wouldn’t have prior exposure to low-level computing.

We created an 8-hour curriculum structured to give both foundational computing and introductory HPC skills to wet-lab scientists. This curriculum was trialled at The Francis Crick Institute to positive feedback. We present this curriculum here as an adaptable framework for other institutions to design a course to empower wet-lab scientists to use HPC resources.

Title: The Careers and Skills for Data‑Driven Research Network+ (CaSDaR)
Authors: Samantha Pearman-Kanza and Louise Saul
Type: Poster
Abstract:
As research becomes increasingly digital and data driven, the sustainability, quality, and reuse of research data depend on more than technical infrastructure alone. Effective digital research practice requires people with the skills, expertise, and capacity to support data creation, management, and reuse across the research lifecycle. Data stewardship has emerged as a way of bringing together these skills and practices, contributing to research capability through support for data quality, documentation, accessibility, and reuse. However, data stewardship roles, responsibilities, and career pathways remain inconsistently defined and supported across institutions and disciplines. Many individuals contribute to this work as part of broader digital research, research software, or technical roles, often without formal recognition, structured development, or clear progression routes.

The Careers and Skills for Data Driven Research Network+ (CaSDaR) is an UKRI funded Network+ established to strengthen capability in this area by building a diverse, inclusive, and sustainable community focused on careers and skills in data stewardship. This poster provides an overview of CaSDaR’s aims, activities, and emerging outputs, including community building, training, skills analysis, and work to improve recognition and career pathways. It situates CaSDaR within the wider ecosystem of national and international initiatives supporting digital research capability and highlights opportunities to engage, collaborate, and contribute to shaping the future of data stewardship as part of sustainable digital research practice.

Title: Identifying Digital Research Bottlenecks and Opportunities in the Age of AI: Insights from an Institutional Survey at the University of Westminster
Authors: Merin Thyagu Pereira, Tamas Kiss, Oba Okekunle, Taiwo Oligbo and Soumya Sharma
Type: Poster
Abstract:
Purpose - The study was conducted as a collaborative work of four Digital Research Technical Champions within the STEP-UP project, which explores digital research challenges and opportunities at the University of Westminster. It aims to inform strategic investment in research infrastructure, tools, and support in an increasingly AI-enabled environment.

Design/methodology/approach – The study adopted a stakeholder-informed approach, including engagement with the Data Accessibility Team, to ensure alignment with institutional priorities in data governance, accessibility, and infrastructure. The study focuses particularly on early-career researchers, highlighting emerging needs within evolving digital research ecosystems. Data were collected via an institutional survey, and the Graduate School supported dissemination via survey links. A total of 24 responses were collected over 15 days. Respondents were predominantly PhD researchers (71%), alongside early-career and senior academics from disciplines including computer science, architecture, linguistics, and life sciences. A mixed-methods approach combined quantitative and qualitative data to capture both trends and lived experiences.

Findings - Findings identify systemic bottlenecks across the research lifecycle, particularly in data analysis, data collection and storage, and data cleaning. Additional challenges include literature management, compliance (e.g., ethics and GDPR), and access to computing infrastructure. While widely used, existing tools lack integration, limiting efficiency and collaboration. Data management practices remain inconsistent and poorly aligned with FAIR principles, reflecting gaps in training, awareness, and infrastructure. Collaboration is further constrained by fragmented platforms and institutional barriers, leading to uneven data accessibility and user inclusivity. Key priorities include AI-driven automation, enhanced data infrastructure, interdisciplinary collaboration platforms, administrative streamlining, and digital skills development.

Research implications for Attendees – The study results would enable attendees to identify comparable bottlenecks in their own contexts and foster discussion on improving data practices, infrastructure, and researcher support to create more efficient, FAIR-aligned, and AI-ready research environments.