Jannik Castenow

Dr. Jannik Castenow

Data Scientist at ITSC GmbH

Helping people live healthier lives through data-driven insights

About

I'm a Data Scientist at ITSC GmbH, building AI solutions for the public health insurance sector. I'm passionate about integrating ML models into traditional workflows — supporting and enhancing the work people already do, not replacing it. My goal is to help people live longer, healthier lives by bringing data-driven insights into the processes that matter.

Before moving into industry, I spent five years as a researcher at Heinz Nixdorf Institute, where I earned my Ph.D. in theoretical computer science and published 13 peer-reviewed papers on distributed algorithms and robot coordination. That academic background gives me a strong foundation for breaking down complex problems and designing rigorous solutions — skills I apply daily when building scalable ML pipelines and RAG-based architectures.

Experience

Data Scientist

Sep 2023 — present

ITSC GmbH

Building AI solutions for public healthcare — integrating ML models into traditional workflows in the health insurance sector.

Research Assistant

Jan 2018 — Jul 2023

Heinz Nixdorf Institute, Paderborn University

Chair of Algorithms & Complexity. Research on distributed computing and mobile robot coordination. Published 13 peer-reviewed papers and presented at international conferences.

Student Research Assistant

Mar 2017 — Dec 2017

Paderborn University

Built a simulator for hybrid communication networks in C++ and conducted research on competitive routing in hybrid networks.

Student Teaching Assistant

Oct 2013 — Feb 2017

Paderborn University

Tutoring for Modellierung, Datenstrukturen und Algorithmen, and Grundlagen der Programmierung 1.

Latest from the Blog

MCP and Claude Code: Under the Hood

Claude Code is an MCP client. It launches servers, runs the handshake, gives the model external tools. The tool call flow, debugging workflow, and building your own server.

  • March 2026
  • MCP
  • Claude Code

Projects

mcptools

mitmproxy for MCP — intercept, inspect, debug, and replay MCP server traffic. Works with Claude Desktop, Cursor, VS Code, and Windsurf.

  • Python
  • MCP
  • DevTools

claude-code-toolkit

Community-curated Claude Code plugins for developer workflows.

  • Shell
  • Claude Code
  • Plugins

nw-migrate

CLI tool that automatically converts Pandas code to Narwhals, enabling backend-agnostic DataFrame workflows.

  • Python
  • Narwhals
  • libcst

Tech Stack

ML & Deep Learning

  • PyTorch
  • scikit-learn
  • XGBoost
  • LightGBM
  • CatBoost
  • Optuna
  • Hugging Face Transformers
  • LangChain
  • SHAP

Data Engineering

  • PySpark
  • Polars
  • Narwhals
  • Pydantic
  • Python

MLOps & Infrastructure

  • Kubeflow
  • Kubeflow Pipelines
  • Docker
  • Kubernetes
  • MLflow

Visualization

  • Matplotlib
  • Seaborn
  • Plotly
  • Streamlit

Interests

Production ML

Building robust, scalable pipelines that take models from experimentation to production. Focused on reproducibility, monitoring, and continuous training workflows.

RAG Architectures

Exploring retrieval-augmented generation systems that combine the strengths of large language models with structured knowledge retrieval for more accurate, grounded outputs.

DataFrame Interop

Working with cross-backend dataframe abstractions like Narwhals to write portable, backend-agnostic data code.

Agentic Coding

Leveraging AI-powered coding agents to accelerate development workflows and explore new paradigms in software engineering.

Conference Talks

Presented research at international peer-reviewed conferences:

  • OPODIS 2022 — Brussels, Belgium
  • SPAA 2022 — Philadelphia, USA
  • SSS 2021 — Virtual
  • SPAA 2020 — Virtual
  • SIROCCO 2020 — Virtual
  • ICDCN 2020 — Kolkata, India
  • ALGOSENSORS 2018 — Helsinki, Finland
  • ALGOCLOUD 2018 — Helsinki, Finland
  • SIROCCO 2017 — Porquerolles, France

Certifications

Deep Learning Specialization

DeepLearning.AI

Professional Scrum Product Owner I

Scrum.org

Professional Scrum Master I

Scrum.org

Education

Ph.D. (Dr. rer. nat.) in Theoretical Computer Science

2018 — 2023

Paderborn University

Dissertation: Local Protocols for Contracting and Expanding Robot Formation Problems

M.Sc. Computer Science

Oct 2015 — Oct 2017

Paderborn University

Semester Abroad

Jul 2016 — Dec 2016

University of Oklahoma

B.Sc. Computer Science

Oct 2012 — Oct 2015

Paderborn University

Publications

  1. Gathering a Euclidean Closed Chain of Robots in Linear Time and Improved Algorithms for Chain-Formation

    Castenow, Harbig, Jung, Knollmann, Meyer auf der Heide

    Theoretical Computer Science 939, 2023

  2. Unifying Gathering Protocols for Swarms of Mobile Robots

    Castenow, Harbig, Meyer auf der Heide

    LNCS, Springer, 2023

  3. A Discrete and Continuous Study of the Max-Chain-Formation Problem

    Castenow, Kling, Knollmann, Meyer auf der Heide

    Information and Computation, 2022

  4. A Unifying Approach to Efficient (Near-)Gathering of Disoriented Robots with Limited Visibility

    Castenow, Harbig, Jung, Kling, Knollmann, Meyer auf der Heide

    OPODIS 2022

  5. The K-Server with Preferences Problem

    Castenow, Feldkord, Knollmann, Malatyali, Meyer auf der Heide

    SPAA 2022

  6. The Max-Line-Formation Problem

    Castenow, Götte, Knollmann, Meyer auf der Heide

    SSS 2021

  7. Gathering Anonymous, Oblivious Robots on a Grid

    Castenow, Fischer, Harbig, Jung, Meyer auf der Heide

    Theoretical Computer Science 815, 2020

  8. The Online Multi-Commodity Facility Location Problem

    Castenow, Feldkord, Knollmann, Malatyali, Meyer auf der Heide

    SPAA 2020

  9. Local Gathering of Mobile Robots in Three Dimensions

    Braun, Castenow, Meyer auf der Heide

    SIROCCO 2020

  10. A Bounding Box Overlay for Competitive Routing in Hybrid Communication Networks

    Castenow, Kolb, Scheideler

    ICDCN 2020

  11. Competitive Routing in Hybrid Communication Networks

    Jung, Kolb, Scheideler, Sundermeier (now Castenow)

    ALGOSENSORS 2018

  12. A Peer-to-Peer Based Cloud Storage Supporting Orthogonal Range Queries of Arbitrary Dimension

    Benter, Knollmann, Meyer auf der Heide, Setzer, Sundermeier (now Castenow)

    ALGOCLOUD 2018

  13. Monitoring of Domain-Related Problems in Distributed Data Streams

    Bemmann, Biermeier, Bürmann, Kemper, Knollmann, Knorr, Kothe, Sundermeier (now Castenow), et al.

    SIROCCO 2017

Contact

Interested in collaborating or just want to say hello? I'd love to hear from you.

mail@janniksu.de LinkedIn GitHub