Blog 2026
Deep Learning Speech Enhancement
A blog series on a denoising autoencoder for speech, from theory to deployment. Improves on reference methods by 69% in PESQ and 9% in STOI.
- Autoencoders
- Signal Processing
- PyTorch
- Deployment
Joining EvoLogics · Nov 2026
Niklas Paulig
Learning systems for the untidy real world.
I work on reinforcement learning, autonomous navigation, and signal processing, with an eye on the constraints that make research useful beyond a clean benchmark.
01
About
My PhD examined reinforcement learning for autonomous vessels on inland waterways, where currents shift, data is noisy, and a controller must contend with traffic rules, shallow banks, and real operating constraints.
I also spent nine months as the founding backend engineer at a mobility startup, building systems for market intelligence and early product development.
From November 2026, I will work as an AI Software Engineer at EvoLogics GmbH.
My notes follow the questions that keep returning: autonomous vessels, signal processing, Fourier transforms, ML deployment, and the occasional useful detour.
Methods & systems
-
Autonomous systems
Decision-making, planning, and control for inland-waterway vessels.
-
Reinforcement learning
Robust policies, simulation, evaluation, and real-world data.
-
Signal processing & deep learning
Denoising, representation learning, and PyTorch workflows.
-
Research software
Python, data pipelines, deployment, and open-source tooling.
02
Selected work
Blog 2026
Visualizing Global Cyber Incidents
An interactive Plotly Dash dashboard on the EuRepoC database: who targets whom, country-level analytics and how the threat landscape has evolved.
- Data Visualization
- Plotly Dash
- Geopolitics
Industry 2025
Nox Mobility · Founding Engineer
Built the technical foundation of a mobility startup from zero to seed, including an automated market-intelligence pipeline for cost forecasts and strategy.
- Python
- Airflow
- PostgreSQL
- Docker
- CI/CD
Research 2025
2-Level RL for Autonomous Vessels
Separate DRL agents for high-level path planning under traffic rules and low-level path following. Beats APF and PID baselines by 65% in obstacle avoidance.
- Deep RL
- Path Planning
- COLAV
Research 2024
AIS Trajectory Extraction (α-method)
An open-source Python framework that extracts long, clean ship trajectories from noisy AIS big data using maneuverability-dependent quantile filtering.
- AIS
- Big Data
- Open Source
Research 2024
Bootstrapped DRL Path Following
A bootstrapped DQN controller for restricted inland waterways such as the Rhine, validated on real-world data and more adaptable than vessel-specific PID control.
- Deep RL
- Path Following
- Real-World Validation
03
Publications
- 2025 2-level reinforcement learning for ships on inland waterways: Path planning and following Martin Waltz, Niklas Paulig, Ostap Okhrin Expert Systems with Applications, Vol. 274 Open Access
- 2024 An open-source framework for data-driven trajectory extraction from AIS data: The α-method Niklas Paulig, Ostap Okhrin Ocean Engineering, Vol. 312 Open Access
- 2024 Robust path following on rivers using bootstrapped reinforcement learning Niklas Paulig, Ostap Okhrin Ocean Engineering, Vol. 298 Open Access
04
Contact
Let's talk research, collaborations, or an interesting ML problem.
The quickest way to reach me is email.