Machine Learning Engineer

Quick Facts

• Focus: training and learning objectives for ML models on sequential/temporal data

• Environment: real data + in-house compute; not primarily prompt engineering or RAG

• Work: experiment design, model training, evaluation, and analysis of what models learn

Description

Build and train your own neural and other ML models over large sequential datasets, with a focus on how the model learns: data representation, prediction targets, architectures, and learning objectives. You will run real training experiments using your own compute infrastructure, analyze training dynamics and failure modes, and propose evaluation experiments to understand why a model learned (or didn’t). The role emphasizes deep ML understanding rather than applying ready-made models only.

Responsibilities

• Design and train neural and other ML models

• Experiment with model architectures

• Work with sequential and temporal data

• Design input representations and define required data and labels

• Propose training objectives and loss functions

• Run experiments to teach specific capabilities

• Support multi-task learning with multiple prediction targets

• Analyze training dynamics, generalization, and failure modes

• Evaluate models and design further experiments

• Diagnose why models learned something (or failed to learn)

Requirements

• Strong ML foundations

• Practical experience with PyTorch, JAX, TensorFlow, or similar ML frameworks

• Understand neural networks, backpropagation, and gradient-based optimization

• Ability to independently design, train, and evaluate ML experiments

• Understanding of how data, labels, architecture, objective function, and evaluation interact

• Ability to convert real problems into well-defined ML/prediction problems

• Experience with sequential or temporal data (or ability to ramp up quickly)

• Understanding of representation learning fundamentals

• Ability to address issues such as class imbalance, sampling bias, noisy labels, and data leakage

• Understanding of generalization and overfitting, including the gap between validation and production behavior

• Ability to plan experiments, compare variants, and use ablation studies rather than random hyperparameter tuning

• Ability to debug an ML pipeline end-to-end (data → labels → representation → architecture → objective → prediction → evaluation)

• Ability to read ML research papers and translate ideas into practical experiments

• Solid foundations in statistics, probability, linear algebra, and optimization

Benefits

• You won’t only use existing models—you’ll focus on how the model learns and what it should learn

• Work with real sequential data and problems where the solution is not predetermined

• Opportunity to design your own architectures, objectives, loss functions, and experiments

• In-house compute enables real training experiments, not limited API-based development

• Influence on research direction, not just implementation

• Deep learning across representation learning, multi-task learning, sequence models, and evaluation

• Research must work beyond benchmarks and notebooks

• Small technical team with high autonomy and direct impact

• Flexible hybrid setup with home work and in-person time in Prague 7