| Verlagslink: | https://ceur-ws.org/Vol-4236/paper1.pdf https://nbn-resolving.org/urn:nbn:de:0074-4236-x http://ceur-ws.org/Vol-4236 |
Titel: | LAMAS : an organisation-centred architecture for agentic AI | Sprache: | Englisch | Autorenschaft: | Dellen, Babette Köhler-Bußmeier, Michael Pakusa, Wied Sudeikat, Jan |
Herausgeber*In: | Köhler-Bußmeier, Michael Moldt, Daniel Rölke, Heiko Capra, Lorenzo Bergenthum, Robin Leemans, Sander Rivkin, Andrey Fahrenberg, Uli Hélouët, Loïc Schlehuber-Caissier, Philipp Ziemiański, Krzysztof |
Schlagwörter: | Organisation; multi-agent systems; self-adaptation; Q-learning; process mining | Erscheinungsdatum: | 2026 | Verlag: | RWTH Aachen | Teil der Schriftenreihe: | PNWS 2026 : joint workshop proceedings of PNSE'26, PNAS'26, ATAED’26, and PHOCON’26 - co-located with the 47th International Conference on Application and Theory of Petri Nets and Concurrency (PETRI NETS 2026) | Zeitschrift oder Schriftenreihe: | CEUR workshop proceedings | Zeitschriftenband: | 4236 | Konferenz: | International Conference on Application and Theory of Petri Nets and Concurrency 2026 | Zusammenfassung: | Current frameworks such as LangGraph, CrewAI, and OpenAI Swarms enable the modelling of static LLM-based multi-agent systems (MAS). However, these systems remain unreliable in dynamic environments due to the lack of clear organisational structures, transparent coordination, and adaptability. The presented project Learn2Organise addresses these limitations by developing an open-source architecture, called Lamas, that supports the creation and execution of learning, self-adaptive, and organisation-centred LLM-MAS. A core idea of Lamas is to complement LLM agents within a symbolically defined organisation to improve explainability and reliability. Lamas does not only support the the specification of agent organisations (with roles and communication flows) and the deployment of LLM-MAS into real-world application processes; it also has a execution engine which enables self-adaptation of organisational structures at run-time through reinforcement learning, including dynamic role assignment and communication path adjustments. Additionally, it also uses process mining techniques to extract interaction and decision structures from log data, automatically generating organisational models and monitoring them at run-time. |
URI: | https://hdl.handle.net/20.500.12738/20355 | ISSN: | 1613-0073 | Begutachtungsstatus: | Diese Version hat ein Peer-Review-Verfahren durchlaufen (Peer Review) | Einrichtung: | Fakultät Informatik und Digitale Gesellschaft Competence Center Smart Systems in Society |
Dokumenttyp: | Konferenzveröffentlichung |
| Enthalten in den Sammlungen: | Publications without full text |
Zur Langanzeige
Volltext ergänzen
Feedback zu diesem Datensatz
Export
Diese Ressource wurde unter folgender Copyright-Bestimmung veröffentlicht: Lizenz von Creative Commons