Rajmund Nagy

Rajmund Nagy

Machine Learning Engineer

KTH Royal Institute of Technology

About me

I am a Machine Learning Master’s student and a former C++ software engineer.

Download my resumé.

  • Machine Learning
  • Generative Modelling
  • Graph Neural Networks
  • Deep Explanation Methods
  • MSC in Machine Learning, 2021 (in progress)

    KTH Royal Institute of Technology, Sweden

  • BSc in Computer Science, 2019

    Eötvös Loránd University, Hungary


Research Engineer (part-time)
KTH, Division of Speech, Music and Hearing
Nov 2020 – Present Stockholm, Sweden

We’re working on a probabilistic gesture generation model with normalizing flows, with a focus on capturing semantic information from in order to generate meaningful gestures.

In collaboration with Taras Kucherenko, Patrik Jonell and prof. Gustav Eje Henter.

Research Engineer (summer internship)
KTH, Division of Robotics, Perception and Learning
Jun 2020 – Aug 2020 Stockholm, Sweden
  • I refactored the source code of Gesticulator, a speech-driven autoregressive neural network for generating gestures.
  • I began working on a framework for integrating similar neural networks into 3D virtual agents (in collaboration with Taras Kucherenko and Ulysses Bernardet)

Supervised by Taras Kucherenko.

Teaching Assistant
Jan 2020 – Nov 2020 Stockholm, Sweden

As part of the TA team, my responsibilities included:

  • Essay grading on the ethical and societal aspects of AI
  • Oral examinations on HMMs and search algorithms in games
  • Testing the assignment code skeletons

The course contents included problem solving with search algorithms, heuristics and games, planning and representing uncertain knowledge and reasoning (Bayesian networks, hidden Markov models).

C++ Software Engineer
evosoft Hungary
May 2018 – Aug 2019 Budapest, Hungary

At evosoft Hungary, I was working in a distributed scrum team on a SCADA platform for Siemens in close collaboration with Austria-based company ETM professional control GmbH.

My responsibilities included:

  • Feature development using C++11/C++14
  • Design and maintenance of integration tests
  • Test reporting and defect analysis for single, redundant, and distributed systems deployed on Windows and Linux (Debian/CentOS)


A framework for integrating gesture generation models into interactive conversational agents.
Analyzing Grad-CAM explanations
We reproduce the original experiments in the Grad-CAM paper, and introduce new evaluation metrics based on adversarial attacks and recent literature.
Analyzing Grad-CAM explanations
Spatial-Temporal Graph Convolutional Network for Action Recognition
Group project for Deep Learning at KTH. A reimplementation of “Spatial-Temporal Graph Convolutional Networks for Action Recognition” (Yan et al., 2018) in PyTorch.
Spatial-Temporal Graph Convolutional Network for Action Recognition