Designing Reliable AI Systems Under Real World Constraints

For decision, communication, and sensing applications

prof_pic_color_square.png

Geronimo Bergk Applied Machine Learning And Edge AI Systems

I work on machine learning systems that remain reliable under real-world constraints such as limited data, energy, and resources. My focus is translating learning algorithms into deployable systems, where robustness, evaluation rigor, and system integration matter more than peak benchmark performance.

I am currently a Managing Consultant in Data Science at Horváth AG, where I design and deploy data-driven decision systems in large-scale enterprise environments. Previously, I was a Research Associate at the Fraunhofer Heinrich Hertz Institute, working on machine-learning-based forecasting, telemetry, and control for communication systems under operational constraints. There, I developed reproducible simulation tools and large-scale datasets for optical networks, integrated learning components into real-time system demonstrators, and contributed to benchmark-driven evaluation pipelines. This work resulted in seven peer-reviewed publications at leading venues and was recognized with the Fraunhofer HHI Emerging Scientist Award for an outstanding master’s thesis.

Engineering principle

I am driven by a system-level perspective that fixes real deployment constraints early, investigates effects isolated in controlled experiments, and judges learning systems by robustness, efficiency, and operability.

Research interests

  • Resource-aware and energy-efficient machine learning for embedded and edge systems
  • Edge AI and TinyML for embedded sensing systems
  • Representation learning under fixed deployment constraints (energy, memory, data)
  • Evaluation protocols, robustness, and reproducibility in real-world ML systems
  • Human-centered sensing with wearables and physiological signals (e.g., biosignals)
  • Edge intelligence for communication-constrained and wireless sensing systems
  • Edge–cloud co-design for long-term, large-scale monitoring applications

Industry practice

  • Design and deployment of data-driven decision systems in finance and operations under strict reliability, auditability, and governance requirements
  • Large-scale forecasting and simulation pipelines operating under uncertainty, incomplete data, and real-time constraints
  • Generative-AI–based reporting systems for executive decision-making, with emphasis on interpretability, traceability, and failure modes
  • Machine learning systems with strong requirements on robustness, interpretability, and organizational accountability

Education

  • M.Sc. Electrical Engineering, 2021 Technische Universität Berlin
  • B.Sc. Electrical Engineering, 2017 Technische Universität Berlin