I'm a computer scientist with a focus on NLP and deep learning. I just finished my M.Sc. at TU Darmstadt, in collaboration with Fraunhofer SIT.
My thesis was on early sexual predator detection in online conversations. I designed a multi-context classifier that combines BERT, GRU and a causal transformer. I also introduced eTSC, a dataset-independent evaluation metric adapted from time series classification. I built the full training and evaluation pipelines myself, with versioned experiments in MLflow and modular code instead of notebooks.
Beyond the thesis, I built ML-based anomaly detection for satellite data in a university project with Telespazio Germany, working in sprints with an aerospace team. At a company hackday I built an RLHF pipeline prototype (SFT, reward model, PPO, DPO) and presented it. I've also built RAG prototypes, one for document retrieval at work and one personal project running locally with LlamaIndex and ChromaDB.
I've worked as a software engineer for nearly seven years alongside my studies (Java/Spring Boot, Angular, Docker, CI/CD). That means I can ship models as production-quality software: tested, containerized and maintainable. I use LLM tooling like Claude Code deliberately in my daily workflow.
I'm looking for remote or hybrid roles in applied ML, NLP or LLM engineering. I'm based in Frankfurt, Germany, and speak English and German at C2 level, plus Italian (native) and French (B2).
Skills
Python, PyTorch, Hugging Face Transformers, BERT, LLMs, RAG, RLHF/DPO, FAISS, scikit-learn, MLflow, Java, TypeScript, Docker, GitLab CI, Linux, PostgreSQL
