Reproducible research code
Researchers learn to let agents add tests, dependency files and documentation to analyses. Results can be rerun by colleagues, reviewers and future group members.
Researchers write a lot of code, but few were trained as software engineers. Our workshop shows research groups, research software engineers, lecturers and campus IT how AI coding agents can make scientific code more reliable instead of just faster.
Project references from our team
+ 15,000 more
In a research group, code is often written by doctoral researchers under deadline pressure, handed over when contracts end and rerun years later for a revision or a follow-up paper. Analyses live in Jupyter notebooks, R scripts or simulation code on a cluster, frequently without tests or documentation. AI coding agents such as Claude Code, OpenAI Codex and Cursor can turn such material into structured packages, add tests and explain what an inherited script actually computes. Used carelessly, they can also introduce subtle errors that change results without anyone noticing.
neooutput, the AI education unit of jocapps GmbH, offers separate tracks for researchers, research software engineers, teaching staff and central IT. Researchers practise reproducible workflows: environment files, tests for numerical results and documentation that a reviewer can follow. Lecturers discuss how agents change programming courses and assessments. Campus IT looks at licensing, data protection and how to provide approved tools to thousands of users. All sessions are available remote or on campus, in English or German.
Researchers learn to let agents add tests, dependency files and documentation to analyses. Results can be rerun by colleagues, reviewers and future group members.
When doctoral researchers leave, agents help newcomers understand inherited code quickly. Groups lose less time reconstructing what a script was supposed to do.
Lecturers get concrete ideas for assignments and exams in a world where students have coding agents. That includes formats that test understanding rather than typing.
Central IT receives criteria for licensing models, data protection and support. That helps to offer approved tools instead of tolerating a patchwork of private accounts.
We talk with the organizers about target groups, disciplines, languages and existing tool policies.
We define separate tracks for researchers, teaching staff and IT, each with suitable examples.
Participants work with agents on their own or sample code, from notebooks to cluster jobs.
Groups receive templates for project instructions, test setups and documentation they can reuse.
Agentic coding workshops and AI services. Hands-on agentic coding workshops for teams on Claude Code, OpenAI Codex and Cursor, remote or on-site, in English or German. Plus AI Code Engineers on demand who deploy, secure and scale AI-built apps.
Explore neooutput workshopsRelated service: Agentic Coding Workshops
No. Some of the biggest gains appear in disciplines where code is a tool rather than the subject, such as life sciences, economics or engineering. We adapt examples to the languages those groups actually use.
Yes, if it contains no protected data and your group agrees. Working on real code makes the session more useful, and we help decide what is suitable beforehand.
We can include job scripts, environment management and workflows that run on shared clusters. Whether agents may run directly on the cluster depends on your computing center's policies.
We discuss assessment formats, transparency rules and course designs that build on agents instead of banning them. The right approach depends on your examination regulations.
15 minutes with an AI Code Engineer, free and without obligation. Or call our team around the clock.