Interdisciplinary Colloquium on Digitalisation of Research: "Research Automation with Agentic LLMs" 04.09.2025 Arman Khalatyan

The presentation of the Interdisciplinary Colloquium on Digitalisation of Research "Research Automation with Agentic LLMs " at 04.09.2025 by Arman Khalatyan

In response to the growing importance of large language models (LLMs) and big data workflows, we have developed a secure, locally hosted AI platform at the Leibniz Institute for Astrophysics Potsdam (AIP). The service, available under https://ai.aip.de, enables all AIP researchers and staff to access advanced LLM capabilities directly—without relying on external providers.
Built entirely on open-source technologies and hosted on AIP’s infrastructure, the platform prioritizes data security, low-latency interaction, and customizability. Users can leverage the system for a wide range of tasks including text generation, code assistance (“vibe coding”), and scientific data workflow generation.
A standout feature is our integration with the REANA reproducible research platform. An in-house agentic LLM assistant can autonomously analyze a scientific query, retrieve datasets, write code, and build a complete REANA workflow pipeline. This pipeline—comprising source code, configuration, plots, and metadata—is executable directly on AIP servers.
This innovation opens new possibilities for AI-assisted science. By automating data analysis tasks and integrating tightly with AIP’s computing ecosystem, ai.aip.de helps researchers focus on scientific discovery while AI handles routine and technical complexity.