Data Scientist

Responsibilities: Design and conduct exploratory data analysis to identify new ideas, hidden patterns, and opportunities for air traffic optimization. As opposed to other domains, the quantity of data in the aeronautical space is not huge, therefore thinking outside of the traditional ML boxes is essential to propose new AI solutions specific to that field. Design, develop, and deploy ML models for real-time classification, regression, and sequence prediction using PyTorch, TensorFlow, or Scikit-learn, including transformer-based architectures for spatial-temporal problems Develop reinforcement learning agents using algorithms such as DQN, PPO, and Actor-Critic, and apply them to simulated and real-world environments via OpenAI Gym or custom setups. Build and optimize RAG pipelines grounded on domain-specific documentation to support AI-generated reasoning and recommendations Evaluate LLM outputs for hallucination and groundedness, and develop domain-specific benchmarks to assess LLM reasoning in ATM contexts Build and automate machine learning pipelines using tools like Kubeflow, Airflow, or similar orchestration frameworks. Design reproducible workflows for data preprocessing, training, evaluation, and deployment. Integrate ML models into scalable APIs and deploy them to cloud-native environments using Docker and Kubernetes. Monitor model performance over time, retrain, and iterate as needed based on live data and production drift. Maintain experiment tracking, model versioning, and reproducibility using tools like MLflow or Weights & Biases. Collaborate with DevOps and backend engineers to ensure seamless integration of ML components into larger systems. Requirements: Education Bachelors in Computer Science or Information Technology Masters degree in Computer Science or Data Science, if applicable Essential Skills/Experience Domain & DataAeronautical domain knowledge would be a major plus. At minimum, some experience in a domain that is not one of the common ones (vision, chatbots...).Experience in a domain where available historical data is not huge would be a plus.Good understanding of data (statistics, features, analytics) and how to map them to the domain.High level of core mathematic skills: algorithms (in particular for prediction), optimizations. Core ML3-5 yrs delivering ML projects end-to-end (data prep to production).Proficiency in Python and machine learning frameworks such as PyTorch or TensorFlow.Familiarity with modern deep learning architectures — including transformers, attention mechanisms, or encoder-decoder models. LLM & RAGExperience building and optimizing RAG pipelines, including chunking strategies, retrieval tuning, and context managementProficiency in prompt engineering and agentic frameworks such as LangChain or LlamaIndexPractical experience evaluating LLM outputs — including hallucination detection, and groundedness against source materialExperience designing domain-specific benchmarks to evaluate LLM reasoning and knowledge in specialized fields Reinforcement LearningSolid understanding of reinforcement learning theory and experience with at least one RL algorithm (e.g. PPO, DQN).Practical experience with OpenAI Gym, Gymnasium, or equivalent RL environments. MLOps & EngineeringStrong grasp of MLOps practices, including the use of Kubeflow Pipelines, MLflow, and model serving platforms.Experience deploying models in containerized environments using Docker and Kubernetes.Hands-on with ETL/ELT tooling (Apache Spark) and modern data-warehouse/lake (S3-based).Knowledge of CI/CD principles for machine learning workflows and model promotion strategies.Ability to build and debug data pipelines that support both training and inference workloads. Desirable Skills/Experience Working knowledge of other languages (e.g., Python3, Scala2 or Scala3, Go, TypeScript, C, C++17, Java17) Familiar with designing and/or implementing AI/MLOps pipelines in public cloud (e.g., Azure, AWS, GCP) Essential / Desirable Traits Possess learning agility, flexibility and pro-activity Comfortable with agile teamwork and user engagement