cv

Research Assistant

Aarhus University, Denmark

Focused on improving the reasoning ability and the cost-quality trade-off of Large Language Models (LLMs). As a first author, two full papers accepted to ICML 2025 and EMNLP 2025. Advisor: Prof. Akhil Arora.

• ICML 2025
• EMNLP 2025

Schuman Trainee

European Parliament

Schuman trainee at the AI unit of the European Parliament. Leveraging LLMs to assess whether an application to the Transparency Register adheres to the applicable regulations.

Student Researcher (Master Thesis)

DLAB, EPFL, Switzerland

Introduced Fleet of Agents, a novel framework utilizing LLMs as agents to navigate through dynamic tree searches. Introduced CacheSaver, a framework that facilitates LLM requests by utilizing asynchronous, batched mechanisms. Perfect grade of 6/6 (top 1%).

• Fleet of Agents
• CacheSaver
• 6/6 grade (top 1%)

Data Science Intern

Prediggo SA, Switzerland

Developed an LLM-powered conversational bot to streamline business rule creation, enhancing user experience and accessibility.

Research Assistant

EPFL FUSTIC, Switzerland

Conducted an extensive literature review and led the data collection and analysis of modern urban platforms, evaluating their capabilities in Urban Digital Twins.

PhD

Aarhus University, Denmark

Computer Science

MSc

Swiss Federal Institute of Technology, Lausanne (EPFL)

Data Science

• Thesis: Fleet of Agents: Coordinated Problem Solving with Large Language Models

BSc

Swiss Federal Institute of Technology, Lausanne (EPFL)

Electrical Engineering

Are Retrials All You Need? Enhancing Large Language Model Reasoning Without Verbalized Feedback

EXAIT Workshop, ICML 2025

Exploring retrials as a method for enhancing LLM reasoning without verbalized feedback.

ReasonBench: Benchmarking the (In)Stability of LLM Reasoning

arXiv preprint, 2025.

Benchmarking the stability and reliability of LLM reasoning capabilities.

CacheSaver: A Modular Framework for Efficient, Affordable, and Reproducible LLM Inference

EMNLP 2025 Findings

A framework that facilitates LLM requests by utilizing asynchronous, batched mechanisms to optimize costs and response time.

Fleet of Agents: Coordinated Problem Solving with Large Language Models

ICML 2025

A novel framework utilizing LLMs as agents to navigate through dynamic tree searches, employing a genetic-type particle filtering approach.

Semester Project - DLAB

Use of foundation models to simulate and study human web navigation.

Deep Learning in Biomedicine - EPFL

Introduced and tested a novel benchmark for few-shot learning algorithms.

Natural Language Processing - EPFL

Used reinforcement learning from human feedback (RLHF) to train an LLM powered chatbot.

Deep Learning - EPFL

Implemented and deployed a deep learning framework to train an image restoration model.