Job Description

Mission

This part-time position (30%) is part of the CROSS 2026 project "Weaponising Culture Online: Detecting and Interpreting Narrative Manipulation in Digital Heritage", a collaboration between EPFL (DHLAB) and UNIL. The project aims to detect and analyze cultural manipulation in Wikipedia and other digital platforms using natural language processing (NLP) techniques and large language models (LLMs).
The assistant will contribute to the development, evaluation, and documentation of a multilingual prototype based on transformer models.

Main duties and responsibilities

Collaborate with other team members to help build the dataset, extracting Wikipedia articles and edit histories to enable annotation and model training. Develop a prototype based on LLMs to detect and classify culturally weaponized content in Wikipedia articles and their revision histories. Fine-tune and evaluate language models (e.g., Hugging Face Transformers) for multiclass classification tasks using the annotated dataset. Design and conduct experiments primarily based on transformer models (e.g., BERT). Document code, ensure reproducibility, and contribute to open-source dissemination of the tool. Work with experts in history and cultural studies to validate model results in a human-in-the-loop setup. Contribute to the writing of technical content for academic publications, project reports, and future research proposals. Participate in EPFL/UNIL research meetings and collaborate within an interdisciplinary team.

Profile

Experience with transformer-based language models (e.g., BERT, RoBERTa, LLaMA) and fine-tuning using frameworks such as Hugging Face and PyTorch. Familiarity with text classification, multilingual datasets, and model evaluation. Strong Python programming skills and experience with version control (e.g., Git). Ability to write clean, reproducible, and well-documented code for research purposes. Interest in digital cultural heritage, or political and historical narratives, is a plus. Excellent command of English (spoken and written) is essential for communication within the project. Knowledge of Ukrainian, Armenian, or Russian is a plus, but not required.

We offer

A stimulating interdisciplinary research environment at the intersection of AI, digital humanities, and cultural studies, within a leading technical university (EPFL) and in collaboration with UNIL. The opportunity to contribute to a high-impact project addressing cultural manipulation and historical narratives in digital spaces. Flexibility in working hours (30% part-time), with the possibility of remote or hybrid work arrangements. Supervision and mentoring by researchers in natural language processing, history, and digital heritage. The chance to co-author academic publications and contribute to open-source research outputs.

Informations

Contract Start Date :01.01.2026

Activity Rate Min : 30 %

Activity Rate Max : 30 %

Contract Type:CDD

Duration: 12 months

Contact : hamest.tamrazyan@epfl.ch