Building and Verifying LLM-Powered Research Software
From research ideas to reliable code: harnessing AI coding agents for reproducible, efficient and sustainable research.
Workshop essentials
30 November 2026
09:00 am – 05:00 pm PKT
SEECS, NUST, Islamabad
H-12 Campus
One-day intensive
Hands-on · bring a laptop
Researchers & practitioners
Limited seats
Agentic AI is transforming how researchers develop, test and maintain scientific software. LLM-powered coding agents can interpret research requirements, navigate codebases, implement algorithms, debug programs and automate development tasks. Their effective use in research, however, demands more than generating functional code: it requires clear task specifications, systematic verification, reproducibility and responsible human oversight.
Agentic AI for Sustainable Innovation is a practical, hands-on training designed to equip researchers and engineers with the skills to brief, supervise and verify AI coding agents. Participants will explore agentic software development workflows, learn to translate research objectives into structured coding tasks, and apply rigorous testing and validation techniques to ensure the reliability of AI-generated research software.
Organised under the UKRI-funded S2Cool (Super-efficient Sustainable Cooling Solution for All Applications) initiative, the workshop connects modern AI-assisted research software engineering with real-world challenges in sustainable cooling, climate analytics, energy efficiency and environmental modelling. Through practical exercises, participants will develop and validate research software components relevant to data-driven sustainability applications.
AI agents can accelerate research software development — but scientific correctness, reproducibility and reliability must remain verifiable.
Researchers, graduate students, software engineers and AI/ML practitioners.
Basic Python programming and familiarity with Git and software development workflows.
A laptop with a configured Python development environment.
By the end of the workshop, participants will be able to:
Explore how LLM coding agents plan, implement, debug and refine research software using development tools and contextual information.
Convert research questions and technical requirements into structured, actionable instructions with clear objectives, constraints, expected outputs and acceptance criteria.
Apply AI-assisted development workflows to scientific data processing, numerical analysis, modelling and AI/ML research tasks.
Design and execute unit tests, integration tests, numerical checks and reproducibility assessments to identify errors, unsupported assumptions and unreliable outputs.
Understand limitations, security risks, data privacy, human oversight and the responsible adoption of coding agents in academic research.
Develop and validate a small research software workflow using climate, cooling or energy-efficiency data.
Seeking to accelerate scientific computing, data analysis, simulation and experimental software development.
Developing research prototypes, machine learning pipelines or computational experiments.
Interested in agentic coding workflows, automated testing and reliable AI-assisted development.
Working on cooling technologies, energy optimisation, environmental modelling and data-driven sustainability solutions.
Responsible for developing, maintaining, testing and reproducing scientific software.
One day, six sessions and a hands-on challenge, 09:00 am to 05:00 pm.
Introduction to LLM coding agents and their role in scientific software development. Agent architectures, tool usage, codebase exploration, planning, execution and human-in-the-loop workflows. Identifying opportunities and limitations in research software development.
Designing effective instructions through structured task briefs, repository context, coding guidelines and acceptance criteria. Hands-on exercise: briefing an AI agent to develop a climate-data analysis or energy-efficiency calculation module.
Hands-on development using LLM coding agents. Implementing data processing pipelines, numerical computations and research utilities. Managing iterative agent interactions, debugging, version control and code improvements, demonstrated on a sustainability-focused task.
Why executable code is not necessarily scientifically correct. Automated unit tests, property-based checks, numerical validation, edge-case testing and reproducibility checks. Identifying hallucinated dependencies, hidden assumptions, logical errors and security vulnerabilities.
Participants use an LLM coding agent to implement and validate a small research pipeline on sample climate or cooling-system data: ingestion, cleaning, cooling and energy-performance indicators, automated testing and reproducible reporting. Each participant prepares an agent brief, reviews the generated code, runs verification tests and documents limitations and corrections.
Responsible use of coding agents in academic research: data protection, code ownership, scientific integrity, verification responsibilities and reproducible research practices. Participant demonstrations, key takeaways and closing.
Participants will leave with:
Organising Team
S2Cool Project Focal Person
SEECS, NUST
Post Doctoral Research Fellow
SEECS, NUST
Trainers
WomeninTechPK
Head of AI
Red Buffer Technologies
Assistant Professor
SEECS, NUST
CTO
LearnOBots
Partners
NUST · UKRI · Ayrton Challenge · Northumbria University UK · WomeninTechPK · LearnOBots · MachVIS
This activity is funded by the UK Government through Project APP47457, titled “Super-efficient Sustainable Cooling Solution for All Applications (S2Cool)”, under the Ayrton Challenge Programme of UK Research and Innovation (UKRI); however, the views expressed do not necessarily reflect the UK Government’s official policies. Project APP47457 is led by Northumbria University, UK, together with its collaborating partners.
Join us to explore how AI coding agents can accelerate research software development while ensuring scientific reliability, reproducibility and responsible innovation.