Projects CV
Back to Projects View on GitHub

Physics-based climate simulation

Climate Simulator

A high-performance, infinite-resolution climate simulation engine. Procedurally generates realistic terrain and simulates physics-based climate dynamics, including sun, temperature, wind, pressure, and water. Numba-accelerated and parallelized, it simulates 60 hours of climate in just one second.

Simulation explorer

Explore the climate fields

Wind

The simulated wind-speed field.

m/s
Wind map at the selected simulation time

Sun

Incoming solar energy.

W/m²
Sun map at the selected simulation time

Pressure

Atmospheric pressure

Pa
Pressure map at the selected simulation time

Air temperature

The simulated temperature of the air.

°C
Air temperature map at the selected simulation time

Soil temperature

The simulated temperature of the ground.

°C
Soil temperature map at the selected simulation time

Water temperature

The temperature of water held on or below the surface.

°C
Water temperature map at the selected simulation time

Condensation

The rate at which water vapour condenses into liquid water.

mm/h
Condensation map at the selected simulation time

Evaporation

Water transferred from the surface into the atmosphere.

mm/h
Evaporation map at the selected simulation time

Precipitation

Rainfall and other precipitation reaching the surface.

mm/h
Precipitation map at the selected simulation time

Water in air

The density of water vapour currently held in the air.

kg/m³
Water in air map at the selected simulation time

Maximum water in air

The maximum water-vapour capacity of the air.

kg/m³
Maximum water in air map at the selected simulation time

Cloud water

Water content suspended in clouds.

kg/m³
Cloud water map at the selected simulation time

Water in soil

The amount of water stored in the soil.

mm
Water in soil map at the selected simulation time
Play 1× 0.25× 0.5× 1× 2× 4× 1 / 225

From terrain to climate

It uses the diamond-square algorithm to generate a 2.5D world terrain, then combines interpolation and high-frequency noise to continuously refine the landscape at arbitrary levels of detail.

Given a latitude and longitude, realistic climate data is used to initialize local atmospheric conditions and solar input. The simulation models the flow of energy from the Sun through the atmosphere and across the terrain, allowing wind and weather patterns to emerge naturally from the underlying physics.

The simulation is designed for high performance, reaching around 60 hours of simulated climate per second on a 256x256 grid.

Diagram of the climate simulator pipeline
Overview of the terrain, energy, atmosphere, and water-cycle interactions.

Berk Can Özmen

+49 xxx xxx xxxx | mail@bcozmen.com | bcozmen | bcozmen.com
Download PDF

Summary

Machine learning engineer and software developer specializing in building end-to-end ML systems and applications. Strong background in Python, PyTorch, deep learning, computer vision, reinforcement learning, and scientific computing.

Skills

Programming: Python, C++, JavaScript, SQL
Machine Learning: PyTorch, Scikit-learn, Deep Learning, Computer Vision, Reinforcement Learning
Sci. Computing: NumPy, SciPy, Pandas, GeoPandas, Numba, CUDA, Matplotlib, Plotly
Backend: Django, FastAPI, REST APIs, PostgreSQL, SQLite, Celery
MLOps: Git, Docker, Linux, Uvicorn, TensorBoard, Testing, Cloud Deployment
Languages: English (C2), German (C1), Turkish (Native)

Experience

06/2025 - Present
Berlin
Freelance ML Engineer
Self-Employed
  • Develop web applications, full-stack machine learning pipelines, and data analysis tools for clients across various industries.
  • bcozmen.com/autocomplete Code and text autocomplete service
    Python · Django · FastAPI · Qwen · Docker · Cloud Deployment
  • bcozmen.com/mlops Full-stack machine learning pipeline for kernel and neural density regression
    Python · Scikit-learn · PyTorch · FastAPI · Docker · PostgreSQL · Cloud Deployment
  • bcozmen.com/rag Wikipedia RAG system
    Python · FastAPI · Qwen · Cloud Deployment
09/2023 - 09/2024
Berlin
Machine Learning Engineer
Fraunhofer SIRIOS
PyTorch · TensorBoard · OSMnx · Geopandas · Wetterdienst
  • Adapted a geospatial emergency response simulation system into a reinforcement learning environment, modeling Berlin's road network, empirical traffic patterns, and incident data.
  • Designed and implemented RL-based control policies that found the optimal solution for analytically solvable cases and produced solutions within emergency response guidelines for complex cases.
  • Designed the reward scheme and evaluation framework to validate policy performance across both scenarios.
09/2021 - 04/2023
Berlin
Computer Vision & Robotics bcozmen.com/scioi
Science of Intelligence
C++ · OpenCV · ROS
  • Collaborated within a research team to develop an autonomous robotic system combining a camera, robotic arm, and soft hand to solve a lockbox puzzle.
  • Implemented a computer vision pipeline that leverages the robotic arm's movement to generate 3D reconstructions and build a model of the environment.
  • Designed feedback loops between the robotic arm and the computer vision pipeline, enabling intelligent exploration of the environment and adaptive interaction based on observations of the lockbox puzzle.
2019 - 2021
Berlin
Internship at Neural Information Processing Group
Technische Universität Berlin
Numba · NumPy · DEAP · neurolib
  • Implemented a simulation environment for FitzHugh-Nagumo based resting-state brain network models, using empirical structural connectivity and functional connectivity data.
  • Explored the parameter space of the FitzHugh-Nagumo model using genetic algorithms and Bayesian optimization with Gaussian processes.
09/2017 - 04/2019
Berlin
Software Developer
Ten8
Selenium · BeautifulSoup · requests
  • Developed and maintained the company's website and internal application for a Berlin-based startup.
  • Developed a web scraping tool to collect and process initial coin offering (ICO) data, automatically alerting the team to new and potentially interesting opportunities.
2013 - 2016
Istanbul
FRC Team Member & Leader
Inanc Mechatronics Club
  • Designed and built competition robots for the FIRST Robotics Competition (FRC), serving as team leader in 2015 and managing the technical team of 10 members.
  • Built robots entirely in-house from design to competition, performing 3D CAD design, welding, machining/milling, mechanical assembly, and programming for both teleoperated and autonomous operation.

Projects

2024 - Present
Deeppy bcozmen/deeppy
PyTorch · TensorBoard
  • Modular deep learning training framework designed for fast prototyping and experimenting
  • Clean separation of model, data, optimizer, scheduler and training logic with automated logging, checkpointing, automatic mixed precision and other efficiency features
  • Clean separation of GPU and CPU space with automatic device placement, memory management and buffer support for large datasets
  • Developed with feedback from practical experience, including my own work and master thesis.
2022 - Present
Agent-Based Neuro-Evolution bcozmen/abne
PyTorch · CUDA
  • Long-term personal research project investigating neuroevolution in agent-based environments without an explicit loss function.
  • Developed a physics-based environment in which agents evolve neural network controllers to survive and adapt to their environment.
  • Investigated low-dimensional parameterizations of agent behavior using message-passing graph neural networks, along with non-destructive crossover and mutation operators for neural network controllers.
  • Aims to understand the emergence of complex behaviors in a multi-agent system and how the selection pressure drives novel strategies that cannot be designed through explicit loss functions.
2025
r-hash-SANE bcozmen/r-hash-SANE
Deeppy
  • A transformer-based rotation-equivariant neural network for learning NeRF latent spaces.
  • Extension of the SANE architecture to hash-grid NeRFs

Education

2026
Berlin
M.Sc. Computer Science, 1.3 (3.8/4.0) bcozmen/master_thesis
Technische Universität Berlin
  • Thesis: Interpreting NeRF Weights Across Architectures with Graph Meta-Networks for Geometry and Appearance Attribution
  • Supervisor: Prof. Dr. Olaf Hellwich
2021
Berlin
B.Sc. Computer Science, 1.7 (3.5/4.0) bcozmen/bsc_thesis
Technische Universität Berlin
  • Modelling Functional Connectivity in Health and Psychiatric Disorders | Thesis Grade: 1.3
  • Supervisor: Prof. Dr. Klaus Obermayer