Internship/Master Thesis (f/m/x) - Multimodal LLM Assistant for Traffic Safety

Short Description of Institute/Facility - Please insert text

Here’s what to expect:
During the development of solutions for the intermodal traffic system, our Institute takes into account the requirements of both the users and the operators. Through the analysis of the required information flows, we support the planning and operation of intermodal overall systems and provide the basis for their technical and operational validation and verification. These systems make it possible to strengthen local public transport and make cities and regions more attractive. For transport hubs such as airports, railway stations or bus stops, we research and optimise infrastructures, processes and the interlinking of all modes of transport. This enables us to increase the safety and predictability of the transport of people and goods.

You will be part of the Algorithms and AI for sensor data processing (AIS) team in the Digitalised Road Traffic (DST) department at the DLR Institute of Transportation Systems. AIS works on a range of projects at the intersection of computer vision, deep learning, multimodal learning and intelligent transportation systems. This internship project focuses on multimodal learning with large language and vision‑language foundation models and their application to traffic safety. This means you will work with state‑of‑the‑art AI methods in a research environment while contributing to solutions that can be deployed in real‑world road traffic systems.

Your responsibilities:
* Analyzing and collecting (if needed) domain-relevant data sources (e.g. traffic regulations, relevant publications, road design guidelines, etc.).
* Development of a dynamic scholarly knowledge base / citation graph of traffic safety and infrastructure publications (paper graph, metadata, citation links).
* Implementing and evaluating an Agentic AI workflow with retrieval-augmented generation (RAG) pipelines in Python to let LLMs answer traffic safety and infrastructure questions with citations.
* Publishing the findings in well-known conferences/journals in the AI and transportation systems space.
* (Optional, depending on your interests) Exploring vision–language models to describe safety issues in images and linking them to literature or rules.

What you bring to the table:
* Ongoing Master studies in e.g. computer science, data science, artificial intelligence, visual computing, or related courses of study.
* Good knowledge of Python, PyTorch/Tensorflow.
* At-least initial experience in the field of computer vision and deep learning (e.g. transformers, LLMs, VLMs).
* Ideally, first experience in at least one of the following areas:

Natural Language Processing or working with LLMs/RAG.
Graph databases / knowledge graphs or network analysis.
Multi-modal large language models (e.g. Gemma4, Qwen)
* Some optional, nice to have skills:

Document processing and conversion (e.g. Docling, Marker)
Document indexing (LlamaIndex, LangChain)
Trust evaluation for LLMs (e.g. Ragas, TruLens)
* Curiosity and intrinsic research motivation to familiarize yourself with complex or new issues.
We look forward to getting to know you!

If you have any questions about this position (Vacancy-ID 5622) please contact:

Dr. Sascha Knake-Langhorst

Tel.: +49 531 295 3474