Dennis Fast

Research Associate, DATEXIS, Berliner Hochschule für Technik (BHT), Berlin, Germany

PhD researcher in AI specialising in medical NLP, evidence-based clinical reasoning and verifiable clinical ML.

About

I work on transformer-based language models for biomedical and clinical text, with a focus on building robust, data-efficient systems that can be applied to real-world medical use cases. My background combines modern deep learning, signal processing and physics-based simulation, which helps me bridge traditional engineering methods and large-scale neural models.

My research aims to build clinical AI systems that do more than generate text - they provide transparent, verifiable and evidence-grounded explanations. I work on developing and evaluating language models and multimodal architectures that can extract clinical evidence, follow structured diagnostic reasoning, and provide clinicians with reliable, interpretable and domain-aware decision support.

I am open to collaborations on verifiable clinical reasoning, evaluation frameworks for medical LLMs, and multimodal clinical AI. If you work on related problems or would like to explore joint research in these areas, feel free to reach out.

Education

  • M.Sc. Data Science (BHT, Berlin): advanced statistics, machine learning methods and deep learning, specialisation in natural language processing (NLP).
  • B.Sc. Applied Mathematics & M.Sc. Computational Engineering (BHT, Berlin): advanced programming skills, numerical methods and scientific computing, specialisation in acoustic simulations.
  • Physical-Technical Assistant (Lise-Meitner-Schule, Berlin): programming skills, experimental methods and applied statistics in science.

Research Interests

  • Transformer models and representation learning for biomedical and clinical text
  • Clinical NLP: information extraction, summarisation and decision support
  • Evaluation and benchmarking of large language models in medical settings
  • Efficient and scalable training of domain-specific language models
  • Physics-based simulation and its intersection with data-driven modelling

Technical Skills & Methods

  • Machine learning and deep learning with PyTorch
  • Pretraining, domain adaptation and fine-tuning of large language models
  • Clinical text processing, terminology handling and outcome prediction
  • Design of benchmarks, evaluation protocols and error analyses for LLMs
  • Multimodal data handling (text, structured EHR data, imaging, audio)
  • Numerical simulation, signal processing and scientific computing

Current Research and Projects

Core Methods & Foundational Model Development
  • Biomedical Language Models for Clinical Text: Development of domain-adapted transformer models for biomedical and clinical documents.
  • Verifiable Clinical Reasoning and Evidence-Grounded LLMs: Designing frameworks for structured, explainable diagnostic reasoning.
  • Reasoning Data Generation and Fine-Tuning of LLMs: Creating reasoning datasets and fine-tuning models for differential diagnosis and evidence-based predictions.
Evaluation & Benchmarking of Medical LLMs
  • Evaluation of Large Language Models for Medical Applications: Benchmarking model accuracy, robustness and reasoning quality on clinical case-based tasks.
  • Benchmarking LLMs for Hepatology: Systematic evaluation of LLMs on liver-disease case data, including guideline adherence and diagnostic precision.
Clinical Applications & Multimodal Extensions
  • COMFORT (Computational Models FOR patienT stratification in urologic cancers – Creating robust and trustworthy multimodal AI for health care): Textual embeddings and downstream models for early detection and prognosis of kidney and prostate cancers. Research is conducted within the EU Horizon project.
  • Beyond STT, End-to-End Audio LLMs for Clinical Workflows: Evaluating audio LLMs for transcription, symptom extraction and respiratory-sound classification; exploring fine-tuning of Qwen3 Omni for explainable audio-based diagnosis.

Collaboration Ecosystem

My current research is carried out within a broad interdisciplinary network spanning radiology, hepatology, oncology, bioinformatics and digital health. I collaborate with clinical, academic and technical partners to ensure that models are developed and evaluated in realistic clinical workflows across multiple medical domains. 

Radiology & Clinical Imaging
  • Technical University of Munich (TUM), Klinikum rechts der Isar – Diagnostic & Interventional Radiology
  • University Hospital RWTH Aachen – Department of Radiology
  • University Hospital Erlangen (FAU) – Department of Radiology
Hepatology & Clinical Bioinformatics
  • Heidelberg University, Medical Faculty Mannheim – Department of Medicine II (Hepatology) and Division of Bioinformatics
  • Center for Preventive Medicine and Digital Health Baden-Württemberg (CPDBW) – Clinical cooperation in metabolic and liver disease research
  • University of Cincinnati, College of Medicine – Division of Internal Medicine and Hepatology
Oncology, Digital Health & Multimodal AI (Audio Project & COMFORT Project)
  • TU Dresden, Else Kröner Fresenius Center for Digital Health (EKFZ) – Digital health research across oncology, internal medicine and computational medicine
  • University Hospital Carl Gustav Carus, TU Dresden – Medical Departments I & III (Oncology and Internal Medicine)
  • German Cancer Consortium (DKTK), Dresden & Mannheim – Clinical AI development and multimodal cancer research
  • University Hospital RWTH Aachen – Diagnostic and Interventional Radiology (clinical imaging integration for multimodal AI)
  • Heidelberg University, Medical Faculty Mannheim – Hepatology and digital health integration for audio-based clinical workflows
Interdisciplinary Themes Across Collaborations
  • Multimodal Cancer AI (COMFORT) – Collaboration across TUM, RWTH Aachen, TU Dresden and Heidelberg University on imaging, text, EHR and biomarker integration.
  • Audio-Based Clinical AI – Collaborative research between hepatology, radiology, digital health and computational linguistics partners for end-to-end audio LLMs in clinical settings.

Publications

Teaching & Supervision

I teach core and advanced topics in NLP for Master’s students at BHT, covering the progression from classical representations to modern architectures. I also teach current advances in LLMs such as reasoning models, agentic systems, and collaborative model workflows.

I also supervise theses on clinical NLP, biomedical language models, clinical reasoning frameworks and applied ML for healthcare. My mentoring focuses on rigorous experimentation, reproducibility and clinically meaningful evaluation. If you have a concrete project idea for a bachalor or master thesis in this area, feel free to get in touch.

List of supervised theses:

Contact & Links

IMG_5407