Ole Behre

M.Sc. Data Science Student | University of Mannheim

Selected Projects

[ 01: Probabilistic Wind Power Forecasting ]

Spatial, probabilistic wind power generation forecasting for Germany. Spatial clustering + LightGBM quantile regression.

-> View Details
[ 02: Bike Traffic Forecasting ]

Comparative analysis of point forecast models on urban mobility signals.

-> View Details
[ 03: Robustness of LLM Reasoning ]

Adversarial stress-testing of LLM robustness. Created novel multilingual, adversarial benchmark.

-> View Details
View on GitHub

Education

Exchange Semester, M.Sc. Data Science Sep 2026 – Jan 2027
National Taiwan University (NTU) | Taipei, Taiwan  [upcoming]
  • Focus: Hardware-proximate AI and sensor-based data acquisition.
  • Courses: Robot Perception and Learning; Mobile and Pervasive Intelligence.
  • Scholarship: Baden-Württemberg-STIPENDIUM by Baden-Württemberg Stiftung.
M.Sc. Data Science 2025 – 2028 (Expected)
University of Mannheim | Grade: 1.5
Focus: Machine Learning, Data Integration, Deep Learning, Large Scale Data Management, Applications of AI in Industry.
B.Sc. Information Systems (Wirtschaftsinformatik) 2021 – 2025
University of Mannheim | Grade: 1.6
  • Thesis: "Multilingual, Adversarial Math Word Problems: Testing the Robustness of Large Language Models" (Grade: 1.0).
  • Relevant Coursework: IT-Security, Artificial Intelligence, Data-Driven Analysis.

Selected Experience

Working Student, Software Engineering Jul 2025 – Present
FORRS | Frankfurt am Main
  • Automate trading-desk workflows and data integrations for short-term power markets, focusing on reliability and operational observability.
  • Contribute to the architecture of a cloud-based market data system (MDM) for ingesting and normalising high-volume energy time-series.
  • Build ETL pipelines and REST APIs to reliably integrate external market data feeds.
Research Assistant Jul 2025 – Oct 2025
University of Mannheim | Data and Web Science Group (DWS)
  • Ran LLM benchmark suites on Slurm-scheduled HPC clusters.
  • Built custom vLLM inference pipelines with intervention hooks to extract and analyse redundancy patterns in the reasoning traces of frontier models (e.g. Deepseek-R1).
  • Conducted quantitative evaluations of inference-time interventions, analyzing the stability and brittleness of Chain-of-Thought (CoT) prompting.
Working Student, Software Engineering (Backend) Sep 2023 – Jan 2025
FORRS | Frankfurt am Main
  • Developed backend services in Java/Quarkus for the MDM cloud platform.
  • Improved REST endpoint performance for high-volume time-series queries.
Working Student, Consulting & Industry Analysis Mar 2022 – Jan 2025
FORRS | Frankfurt am Main
  • Conducted data analysis and market research across the energy trading value chain to support consulting projects.
  • Produced and organized a targeted podcast series exploring the technical and business intersection of Data Science and the energy industry.
Member Mar 2023 – Present
STADS e.V. (Student Association for Data Analytics)
  • Website maintenance and infrastructure restructuring.
  • Automation of internal association processes.
  • Trying to get people to join the STADS Running Club ;)

Technical Specifications

Languages Python, Java, SQL, HTML/CSS
ML & Forecasting scikit-learn, LightGBM, CatBoost, pandas, NumPy, conformal prediction, time-series modeling
LLMs & Deep Learning PyTorch, Transformers, vLLM, lm-eval, Hugging Face, Slurm/HPC
Backend & APIs Quarkus, FastAPI, REST, WebSockets, PostgreSQL, ETL pipelines
Infrastructure Docker, Linux, Git (GitHub, GitLab)
Ways of Working AI-native development (agentic coding tools, LLM-assisted workflows), Agile/Scrum, Jira, Confluence, cross-functional collaboration
Spoken Languages German (native), English (C2 equivalent), Mandarin Chinese (Currently Learning!)

Deep Dives

[+]
In-Context Forecasting in Supply Chains: Evaluating the Promise and Limits of Tabular Foundation Models
TabPFN Demand Forecasting In-Context Learning
| 2026

Evaluating the limits and promise of Tabular Prior-Data Fitted Networks (TabPFN) and ApolloPFN for zero-shot demand forecasting in data-scarce supply chains, highlighting trend extrapolation and context window bottlenecks.

-> Read Paper (PDF)
[+]
Deep Learning for Bearing Predictive Maintenance: A Review of Four Different Architectures
Deep Learning Predictive Maintenance LSTM
| 2026

Critically analyzing four deep learning approaches to bearing remaining useful life (RUL) prediction (DNN, Stacked Denoising Autoencoder, CNN-based transfer learning, and LSTM-fusion) and uncovering structural data leakage issues in the baseline evaluation.

-> Read Paper (PDF)
[+]
Red Teaming LLMs: Gender Bias in AI-Generated Parenting Advice
LLMs Red Teaming Gender Bias
| 2026

Auditing implicit gender bias in frontier LLMs within the parenting and child development advice domain using identity swapping red teaming to investigate allocative and representational harms.

-> Read Paper (PDF)

Off-Screen

Bikepacking, Bouldering & Climbing, Casual Running, Specialty Coffee, Video Games.

Here for fun?

[ 01: Grid Control ]

A grid balancing and dispatch mini-game. Manage variable renewables, industrial storage, and carbon guilt. ʕ •ᴥ•ʔ

-> Enter Dispatch Center

Links & Contact

[ querying live weather data... ]
Animation of a wind park representing renewable energy interest.