AI Engineer – Decision Making
Location: Berlin, Germany | Employment: Full-time
Reporting to: VP | Start: As soon as possible

Build the intelligence that helps autonomous vehicles decide what to do next.

DeepRec is partnering with a German autonomous driving company developing a full Level 4 system for European roads. Combining perception, explainable AI, decision-making and vehicle control, the team is building technology designed to make autonomous mobility safe, understandable and ready for certification.

We’re looking for an AI Engineer to develop and improve deep learning models for autonomous decision-making and trajectory planning. You’ll join a small team bringing together expertise in reinforcement learning, active inference, deep learning and deployment.

The role

Your focus will be practical experimentation: implementing model architectures, training and fine-tuning models, reproducing research results and using data to understand what improves performance.

A key challenge is improving closed-loop behaviour—how the vehicle performs when a generated plan is executed by the controller and the resulting feedback informs the next decision. You’ll work closely with simulation and other engineering teams to evaluate this interaction and translate findings into better models.

Strong hands-on deep learning experience is essential. You should be comfortable moving independently from a research paper or experimental hypothesis to a working implementation, evaluating the results and deciding what to change next.

What you’ll do

  • Design, train and fine-tune deep learning models for autonomous driving decision-making and trajectory planning.

  • Translate research papers into working implementations, reproduce experimental results and adapt promising approaches to the vehicle’s requirements.

  • Run structured experiments, tune model parameters and compare results against established metrics and benchmarks.

  • Analyse datasets and model failures to identify gaps in coverage, improve training inputs and strengthen model performance.

  • Work with simulation and control colleagues to evaluate closed-loop behaviour and improve the interaction between planning, execution and feedback.

  • Investigate performance differences between offline datasets, simulation and real-world environments.

  • Collaborate with data, perception and simulation teams to ensure models use relevant data and are tested against meaningful driving scenarios.

  • Support model integration into the wider autonomy stack and, as the platform develops, optimise for latency, memory and compute constraints.

What you’ll bring

  • Strong practical experience in deep learning, typically gained through three or more years of relevant work, with the depth of your contribution valued over a rigid tenure requirement.

  • Strong Python skills and hands-on PyTorch experience.

  • A solid understanding of model architectures, training dynamics, optimisation and evaluation.

  • The ability to implement research independently, reproduce results and systematically investigate why an experiment succeeds or fails.

  • Experience applying learning-based methods to trajectory planning, decision-making or closely related autonomous systems problems.

  • The ability to use data analysis to improve models, including identifying failure patterns and making informed changes to datasets, training or evaluation.

  • An understanding of generalisation challenges when moving between datasets, simulation and real-world conditions.

  • An open-minded, experimental approach and the ability to collaborate closely with engineers across disciplines.

Automotive or autonomous driving experience is particularly valuable; relevant robotics experience is also welcomed. Previous production deployment experience is useful, but not essential if you bring strong model-development and experimentation skills.

Useful additional experience

  • Reinforcement learning or active inference.

  • Autonomous driving simulation environments such as CARLA.

  • Experiment tracking with tools such as Weights & Biases or MLflow.

  • ROS/ROS 2 and C++ integration.

  • Model deployment and optimisation using ONNX, TensorRT or quantisation.

  • Training in GPU compute environments and working with Git, Docker and CI workflows.

Why join?

You’ll work on a central challenge in autonomous driving: turning learned models into decisions that perform reliably as the vehicle interacts with its environment.

In a small, multidisciplinary team, you’ll have direct influence over experiments, model choices and evaluation methods. The immediate opportunity is to improve closed-loop performance, with further scope to tackle efficient inference on constrained vehicle hardware as the platform matures.

Interested? Contact DeepRec to discuss the role confidentially.