Master Thesis : Explore advanced memory patterns to improve agentic LLM applications

Description
The current generation of agents deployed in a company rarely rely on a structured memory of previous interactions. However, as agents become prevalent, making sure that they consider user preferences in a dynamic way will become more important.

This is especially true in the context of a central routing agent, to make sure that the context of the user is taken into consideration to better serve them.

It however raises several interesting challenges that will have to be investigated in this project.

For instance, can the memory adapt to a moving context? A user that started a new position has no longer the same centers of interests. Or simply, if a user now generally wants more verbose answers whereas it preferred before terse answers with simple links.

Storing this information for a company would also raise questions of confidentiality, and a local setup will need to be used. It will be important to let users access the stored memories and have full control over them.

To explore this theme, we will focus on an agent architecture that will handle centrally the memory, and rely on secondary agents performing RAG on specific parts of ELCA’s knowledge base

It will benefit from the work done currently at ELCA on developing multi-agent infrastructure with automatic discovery.

The objective would be to demonstrate the practical value of this approach and confirm that the cost-to-reward ratio justifies its use in real-world business application.

This internship offers a rich ground for hands-on experience, particularly in a research area aiming to expand the operational scope of conversational AI through leveraging proprietary data and the relatively unexplored field of memory in multi-agent LLMs. This is both an educational and professional development opportunity in a rapidly evolving technological landscape.

Objectives

• To explore and deploy LLM memory frameworks (e.g. Mem0) and create a program around it to make it useful in a business context.
• To deliver a proof-of-concept application capable of leveraging this memory-based agentic approach, and evaluate its usefulness.
• To develop real hands-on experience on currently used chatbots

Our offer

• A dynamic work and collaborative environment with a highly motivated multi-cultural and international sites team
• The chance to make a difference in peoples’ life by building innovative solutions
• Various internal coding events (Hackathon, Brownbags), see our technical blog
• Monthly After-Works organized per locations

Skills required

• Experience with Machine Learning and NLP principles, familiarity with LLMs.
• Experience with Python (Pandas, PyTorch, …) and software engineering principles.
• Web development skills are appreciated (React, Streamlit, …).
• Strong communication skills in both French and English, capable of articulating complex ideas clearly in both written and verbal forms.
   

This internship starts in February 2026. 

Applications must include your most recent academic transcripts (grades); applications without transcripts will not be considered.