Agentic AI for Biomedical Text Generation - Master's Thesis / Internship @ Helmholtz Munich

02.09.2026, Studentische Hilfskräfte, Praktikantenstellen, Studienarbeiten

Project Background

A Master's internship at Helmholtz Munich in the Lobentanzer Lab, building on ongoing project work with the German Center for Diabetes Research (DZD) and the Federal Institute of Public Health (BIÖG), sitting at the intersection of natural language processing and public health communication.

Automatically generated summaries are increasingly used to communicate medical information and findings to lay audiences, but standard surface-level metrics often fail to capture whether generated text still means what the source document meant. This project builds a dataset of source documents paired with automatically generated summaries, and develops evaluation approaches that go beyond surface-level metrics to detect systematic information distortion. The goal is to develop reusable evaluation methodology for high-stakes text generation, not to benchmark specific models.

Your Tasks

• Design and build a structured dataset pipeline pairing source documents with automatically generated summaries.
• Run experiments across multiple language models and generation strategies.
• Investigate how and where information distortion occurs between source and generated text, and whether it can be detected systematically.
• Work with large language models, embedding models, and multi-agent pipelines.
• Build robust logging and experiment tracking infrastructure.

Project Details

Prerequisites: Strong background in NLP and machine learning; solid Python skills; hands-on experience with LLMs (e.g., via HuggingFace or OpenRouter) and embedding models. Familiarity with evaluation methods for generated text or biomedical/clinical text is a plus.

Preferred Start Date: As soon as possible (flexible, to be agreed individually).

How to Apply: Please send a CV, a brief motivation letter (max. one page), and a transcript of records to raeesa.yousaf@helmholtz-munich.de

Kontakt: raeesa.yousaf@helmholtz-munich.de

Mehr Information

https://www.slolab.ai/