Norsk English

Generative AI to Enable Multi-Agent Conversations

The thesis explores multi-agent conversation and addresses its challenges. Large Language Model (LLM)-based agents are entities that harness conversational capabilities, designed to solve specific tasks, such as text or code generation.

Contact persons

Ill.: SINTEF

Master Thesis Description

In the digital world, interactions are often characterized by predefined workflows and lack of semantic comprehension. Multi-agent conversations can improve or replace these interactions, allowing systems to collaborate in a human-like manner.  The thesis objective is to enhance interoperability by implementing the actor model, exploring the interaction dynamics among multiple agents (or actors), and addressing their limitations.

Research Topic Focus

  • Analysing the state-of-the-art in multi-agent conversation systems.
  • Developing a framework that adheres to the actor model guidelines, facilitating effective communication and coordination between agents.
  • Use of open source LLMs as part of agent creation.
  • Implementing and testing the framework within simulated or real edge-cloud continuum environments.

 Expected Results

  • A comprehensive analysis of the challenges and opportunities in current system communication and multi-agent conversation systems.
  • A novel multi-agent framework aligned with the actor model principles, enabling efficient and reliable collaboration between LLM-based agents.
  • Demonstrated improvements in the collaborative efficiency of multi-agent conversations compared to state-of-the-art.
  • Insights into the applicability of the proposed solutions across various use cases.

Learning Outcomes

  • Acquire hands-on experience in designing and implementing advanced conversational agents and interaction strategies.
  • Gain a better understanding of the technological underpinnings of generative AI, LMMs and multi-agent conversation systems.
  • Improve ability to critically assess the performance and effectiveness of generative AI-enhanced systems.
  • Innovate the field of AI.

Qualifications

  • Strong background in machine learning, AI and LLMs.
  • Proficiency in programming languages commonly used in AI development, with a preference for Python.
  • Analytical and creative thinking.

References

Wu, Q., Bansal, G., Zhang, J., Wu, Y., Zhang, S., Zhu, E., ... & Wang, C. (2023). Autogen: Enabling next-gen LLM applications via multi-agent conversation framework. arXiv preprint arXiv:2308.08155.

Josifoski, M., Klein, L., Peyrard, M., Li, Y., Geng, S., Schnitzler, J. P., ... & West, R. (2023). Flows: Building blocks of reasoning and collaborating AI. arXiv preprint arXiv:2308.01285.

Hewitt, C. (2010). Actor model of computation: scalable robust information systems. arXiv preprint arXiv:1008.1459.