Software Engineering in Python

Workshop at FHI, March 23-24, 2026

AI and Data Analytics Group

  • Piero Coronica     piero.coronica@mpcdf.mpg.de
  • Nastassya Horlava     nastassya.horlava@mpcdf.mpg.de

Application Support Group

  • Tobias Melson     tobias.melson@mpcdf.mpg.de

Monday

  1. Introduction
  2. Code structure
  3. Organize the codebase
  4. Git for code versioning

Tuesday

  1. Open source publishing
  2. Testing
  3. Project lifecycle
  4. Python script parametrization
  5. Git for collaborative projects
AI and HPDA Support @ MPCDF

AI and HPDA Support @ MPCDF

Since 2018 a dedicated team has been established to support MPG researchers with their AI/HPDA workloads on MPCDF systems and to deliver dedicated project support.

Currently team of 5 people (various backgrounds, all strong knowledge about AI/ML and HPC):

  • Timoteo Colnaghi (Physics)
  • Piero Coronica (Mathematics)
  • Nastassya Horlava (Computer Science)
  • Andreas Marek (Physics)
  • Christina Winkler (Computer Science)

How to reach us:
andreas.marek@mpcdf.mpg.de or support@mpcdf.mpg.de

AI and HPDA Support @ MPCDF

Current and past collaborations with MPIs (over 14 institutes)

  • Animal Behavior (video analysis)
  • Astronomy (Gaussian process optimization)
  • BIGmax (general AI support)
  • Biology of Aging (Spark cluster infrastructure)
  • Brain Research (code optimizations)
  • CBS (Image Segmentation/Reconstruction, Super Resolution)
  • Collective Goods (RAG systems)
  • Colloids and Interface: 2d/3d image segmentation
  • FHI (parallel training, GANs)
  • Legal History and Legal Theory: LLMs
  • Microstructure of Physics: image segmentation
  • Physics: surrogate models
  • Plasma Physics: parallel training, HPO
  • Sustainable Materials: special DL models for material classification, LLM finetuning
AI and HPDA Support @ MPCDF

Some examples of AI usage at MPCDF

Over 17 larger projects from 12 MPIs supported in the past few years

Retrieval Augmented Generation

Recognition of Crystal Structures

Surrogate models & error correction

AI and HPDA Support @ MPCDF

NVIDIA Deep Learning Institute (DLI) at MPCDF

Certified instructors for:

  • Fundamentals of Deep Learning
  • Data Parallelism: How to Train Deep Learning Models on Multiple GPUs
  • Applications of AI for Anomaly Detection
  • Building Transformer-Based Natural Language Processing
  • Adding New Knowledge to LLMs

Next certifications:

  • Model Parallelism: Building and Deploying Large Neural Networks
Application Support @ MPCDF

Application Support @ MPCDF

Team of currently 16 HPC experts with various scientific background

  • Software support for users on MPCDF systems: compilers, libraries, tools
  • Performance analysis and code optimization for HPC applications, not limited to MPCDF hardware
  • Many long-term projects in close collaboration with MPG scientists
  • Scientific visualization: software and project support
  • Assessment of novel hardware and software: close collaboration with hardware vendors
  • Training: lectures, tutorials, workshops