PPAM 202630 Aug.–2 Sept.
Poznań © Diego Delso / CC BY-SA 4.0
Onedata at PPAM 2026: LLM Agents and FAIR Data in Federated Research Infrastructures
- conference
- Name
- 16th International Conference on Parallel Processing and Applied Mathematics (PPAM 2026)
- When
- 30 August – 2 September 2026
- Where
- Poznan University of Technology, Poznań, Poland
The 16th International Conference on Parallel Processing and Applied Mathematics (PPAM 2026) took place at Poznan University of Technology from 30 August to 2 September 2026. The conference brought together researchers working on parallel processing, distributed computing and their applications.
Onedata contributed to the programme with two papers addressing complementary challenges for federated research data infrastructures: enabling LLM agents to interact with scientific data and supporting a FAIR, interoperable and machine-actionable data lifecycle.
LLM-agentic access to a federated scientific data layer with Onedata
On Monday, 31 August, the team presented “LLM-agentic access to a federated scientific data layer with Onedata” as part of Track E: Main Track – Environments for Parallel and Distributed Computing.
The presentation explored how LLM agents can interact with federated scientific data through Onedata using a curated 16-tool MCP server. The approach was evaluated in more than 1,000 trials across seven independently trained LLMs on a live multi-provider federation, achieving a 94.8% pass rate and revealing MCP-server design issues that could remain hidden in single-model evaluations.
The paper was authored by Michał Orzechowski, Renata G. Słota, Mikołaj Maślak, Łukasz Dutka and Jacek Kitowski.
Monday, 31 August 2026, 16:35–17:00 CEST
Track E: Main Track – Environments for Parallel and Distributed Computing
Paper: LLM-agentic access to a federated scientific data layer with Onedata
Authors: Michał Orzechowski, Renata G. Słota, Mikołaj Maślak, Łukasz Dutka, Jacek Kitowski
FDIM: FAIR Data Integration Model for Federated Data Infrastructures
On Tuesday, 1 September, the team presented “FDIM: FAIR Data Integration Model for Federated Data Infrastructures” during Track E: Workshop on Advances of Global Challenges Applications.
The paper introduced FDIM, a lightweight, infrastructure-independent model for supporting the FAIR data lifecycle across large federated dataset collections without modifying the underlying storage system.
The model combines metadata normalisation, lifecycle and provenance management, and machine-actionable access. A proof-of-concept implementation in Onedata achieved near-complete FAIR assessment scores, while its hierarchical linkset design enabled large collections to be crawled more efficiently than a flat baseline while maintaining constant memory usage.
Łukasz Opioła co-authored the paper with Adrianna Bodziony, Adam Biśta, Łukasz Dutka, Renata G. Słota and Jacek Kitowski.
Tuesday, 1 September 2026, 17:30–17:55 CEST
Track E: Workshop on Advances of Global Challenges Applications
Paper: FDIM: FAIR Data Integration Model for Federated Data Infrastructures
Authors: Adrianna Bodziony, Adam Biśta, Łukasz Opioła, Łukasz Dutka, Renata G. Słota, Jacek Kitowski