29.09.2026 aktualisiert
nicht verfügbar
info: nicht verfügbar bis 31.12.2026
Deep Learning Researcher & Applied ML Engineer
Mainz, Deutschland
Mainz +50 km
PhD in Deep Learning, Johannes Gutenberg-Universität Mainz (Magna Cum Laude, Grade 1.0)Über mich
Deep Learning Researcher and ML Engineer with 8+ years of experience, a PhD, and first-author ICML/ICLR publications. Expertise in neural architectures, training dynamics, optimization, and efficiency, plus industry experience in LLM post-training, AI agents, evaluation, and synthetic data.
Skills
AI AgentsLLMFine-TuningKünstliche IntelligenzPythonMachine LearningData SciencePyTorchLarge Language ModelsDeep LearningGenerative AI
Neural-Network Architecture Design & Research
Expertise in neural-network architecture design, training dynamics, optimization stability, and model performance in deep neural networks, backed by first-author publications at ICML and ICLR.
LLM Fine-Tuning, Distillation & Applied AI Pipelines
Design and implementation of custom ML/AI pipelines including LLM fine-tuning and distillation, LoRA, encoders and rerankers, agentic systems, tool use and orchestration, recursive self-improvement, and context engineering.
Model Compression & Efficiency
Research and application of model compression techniques, including extraction of shallow subnetworks and reduction of inference cost by orders of magnitude while maintaining near-frontier-model performance.
Synthetic Data Generation
Design of synthetic-data generation pipelines enabling models trained on generated data to transfer successfully to previously unseen real-world enterprise environments.
Rigorous Evaluation Methodology
Construction of rigorous evaluation infrastructure with frozen holdouts, pre-defined criteria, provenance and contamination controls, and ablation testing, including identification of label leakage and establishment of corrected results.
Self-Improving Training Loops
Development of self-improving training loops using adversarial data generation and carefully designed acceptance criteria, establishing transferable performance gains on previously unseen holdout data.
Large-Scale ML Pipeline Optimization
Building and optimizing large-scale ML pipelines, including training GNNs on internet-scale telemetry and reducing training workflows from 9 hours to 30 minutes through batch-size optimization, learning-rate tuning, and component ablation.
Tech Stack: Python, PyTorch & Cloud Infrastructure
Proficiency in Python, PyTorch, Hugging Face/Transformers, scikit-learn, SQL, C++, Git, Linux, Docker, CI/CD, GCP, AWS, and SLURM.
Mathematical Modeling & Optimization
Strong mathematical background in calculus, statistics, probability, numerical methods, and mathematical data modeling, with specialization in optimization and machine learning.
Expertise in neural-network architecture design, training dynamics, optimization stability, and model performance in deep neural networks, backed by first-author publications at ICML and ICLR.
LLM Fine-Tuning, Distillation & Applied AI Pipelines
Design and implementation of custom ML/AI pipelines including LLM fine-tuning and distillation, LoRA, encoders and rerankers, agentic systems, tool use and orchestration, recursive self-improvement, and context engineering.
Model Compression & Efficiency
Research and application of model compression techniques, including extraction of shallow subnetworks and reduction of inference cost by orders of magnitude while maintaining near-frontier-model performance.
Synthetic Data Generation
Design of synthetic-data generation pipelines enabling models trained on generated data to transfer successfully to previously unseen real-world enterprise environments.
Rigorous Evaluation Methodology
Construction of rigorous evaluation infrastructure with frozen holdouts, pre-defined criteria, provenance and contamination controls, and ablation testing, including identification of label leakage and establishment of corrected results.
Self-Improving Training Loops
Development of self-improving training loops using adversarial data generation and carefully designed acceptance criteria, establishing transferable performance gains on previously unseen holdout data.
Large-Scale ML Pipeline Optimization
Building and optimizing large-scale ML pipelines, including training GNNs on internet-scale telemetry and reducing training workflows from 9 hours to 30 minutes through batch-size optimization, learning-rate tuning, and component ablation.
Tech Stack: Python, PyTorch & Cloud Infrastructure
Proficiency in Python, PyTorch, Hugging Face/Transformers, scikit-learn, SQL, C++, Git, Linux, Docker, CI/CD, GCP, AWS, and SLURM.
Mathematical Modeling & Optimization
Strong mathematical background in calculus, statistics, probability, numerical methods, and mathematical data modeling, with specialization in optimization and machine learning.
Erfahrung
Berufserfahrung5 - 10 Jahre
Freelancer seit2026
Sprachen
DeutschMutterspracheEnglischverhandlungssicherFranzösischMutterspracheItalienischgutSpanischgut
Projekthistorie
Seed-stage AI-security startup building an agentic SOC platform. Designed a synthetic-data generation pipeline for transfer to unseen enterprise environments. Built an ML pipeline combining classical ML with fine-tuned specialist LLM components for near-frontier performance at reduced inference cost. Built rigorous evaluation infrastructure and identified label leakage. Developed self-improving training loops using adversarial data generation. Measured the value of structured context for LLM agents to improve threat-detection performance. Built and optimized large-scale ML pipelines including GNN training on internet-scale telemetry.
Dissertation on how specific architectural choices (e.g., batch normalization, residual connections) affect optimization stability, robustness, and model performance in deep neural networks. PhD completed in January 2025 with Magna Cum Laude, Grade 1.0. Continued as a postdoctoral researcher through September 2025 to finalize ongoing research. First-author publications at ICML and ICLR. 8+ years of university-level teaching experience from undergraduate to advanced graduate-level courses. Reviewer for NeurIPS 2025.