TPC Hackathon: Improving AI for Science

In the light of the rapid global spread of generative AI tools and the risks they pose, the TPC was launched during the SC23 supercomputing conference by ANL, the Japanese RIKEN Centre and European Barcelona Supercomputing Centre (BSC). “TPC aims to become a sustainable movement for the use of AI in science and to promote research into it, as well as trustworthy programmes,” explains Catlett. In addition to regular hackathons, TPC organises conferences and workshops worldwide and publishes tutorials on AI-supported research.

From the outset, European high-performance computing centres such as the LRZ have actively supported this community. Consequently, TPC Europe was established in Barcelona in 2024, its own office and organisational structures followed in spring 2026 with the funding of the EuroTPC Project by EuroHPC Joint Undertaking: During the TPC annual conference 2026 in Boston in early June, members from Europe met at the LRZ. Some of their presentations were streamed to the US and to other satellite events, and vice versa. Two of the four TPC hackathons of the year took place in Europe, in spring at CINECA in Italy, in fall at LRZ in Garching. “Within the EuroTPC Project, the LRZ has taken on the task of drawing up a EuroTPC roadmap,” reports Dr. Nicolay Hammer, head of the LRZ’s Big Data & AI (BDAI) team. “To this end, we are currently gathering recommendations and topics from researchers and institutes to devise EU strategies and foster the use of AI in science.”

Agentic AI will feature very prominently as a topic in this context. Special tasks were already suggested for the TPC hackathon in the context of agentic use of AI workflows. In those cases large language models coordinate jobs in data processing together with other AI tools and interact with them to carry out actions. 

For example, one of the teams at LRZ worked on agentic workflows to develop hypothesis generation for protein. Another working group from the Gauss Supercomputing Centre (GCS) developed initial workflows to enable an AI agent to adapt code to various supercomputers, thereby simplifying access to a wide range of HPC systems. “For me, the most important thing during the TPC hackathons is the exchange of ideas; it also allows me to learn about different approaches or how technology is used,” says Mitja Sainio, a Machine Learning (ML) engineer at the Finnish CSC – IT for Science Centre, who collaborated with the GCS team. In addition, the hands-on work helps building networks of contacts that AI specialists can turn to for practical advice in their day-to-day work.