Stellenticket Humboldt-Universität zu Berlin

Supporting activities under guidance in the “Text2DSL” project:

1. Contribute to the R&D of a Text2DSL co-pilot for enhancing domain specific code generation capabilities using Large Language Models (LLMs) (50%)
2. Assist in designing and structuring datasets, fine-tuning LLM models, and defining key performance indicators for experiments (25%)
3. Support with the implementation both intrinsic and extrinsic evaluations of the Text2DSL co-pilot, including the identification and application of relevant metrics (25%)

The place of work (on-site) is the TU Campus Charlottenburg in Berlin.

Must:

1. Experience in fine-tuning open-source Large Language Models (LLMs)
2. Good knowledge in the Hugging Face Transformers ecosystem
3. Enrolment at a German University
4. Good knowledge of German and/or English; willingness to learn any missing language skills

Can:

1. Basic understanding of Retrieval Augmented Generation (RAG) and AI Agentic approach

Please submit your application documents summarized in one PDF.

Party responsible for specialist area / point of contact for job posting: Adnan Ahmad, Frau Irene Hube-Achter
Period of employment: immediately until 31.12.2026
Apply to: bewerbung@qu.tu-berlin.de

Please submit your written application including cover letter, your CV, certificate of enrollment, and where applicable, current transcript of records, with the reference number to the place of employment indicated above.
In the interest of promoting equality opportunities for men and women, applications from women with suitable qualifications are particularly encouraged.