Sign inGet app

Jobs › Listing 3d ago

Applied ML Engineer // LANIAKEA solutions

LANIAKEA Solutions · Berlin, Germany

Full-timeOn-siteEnglish

About the role

From the employer’s listing · LANIAKEA Solutions · posted 17 November 2026

Applied ML Engineer: Time Series and Optimization (Industrial)

We are looking for an Applied ML Engineer focused on time series and optimization in an industrial setting. The role is very hands-on: models have to run on-premise and be explainable.

Responsibilities

  • Develop ML models that correct and complement physical and rule-based models.
  • Build load and consumption forecasts on sensor time series.
  • Integrate models into an existing optimization system and run them on-premise.
  • Set up training data, validation, and drift detection.
  • Give traceable reasons for the models’ recommendations.
  • Prepare for reinforcement learning once enough operational data is available.

Required skills

  • Time series ML on noisy, incomplete sensor data.
  • Hybrid modeling: ML as a corrector on top of physics and rule-based models.
  • Understanding of linear and mixed-integer optimization.
  • Explainability methods and the ability to convey them to non-ML experts.
  • MLOps without cloud: versioning, drift detection, retraining.
  • Production-grade Python with tests and type annotations.
  • At least one ML model personally taken into production.

Preferred skills

  • Reinforcement learning, especially offline RL and safe RL.
  • Domain knowledge in thermodynamics, energy engineering, or control engineering.
  • Experience with industrial data connectivity (MQTT, OPC UA) and edge inference.

Skills mentioned

mlinferencePython

See your match score for every role

BabZituna scores every job against your profile on six real dimensions and shows you WHY it scored that way, audited for fairness (read the public bias audit).

Get the app → ✓ 100% free for job seekers
How one job scored Example
Skills96Experience90Location84Work style74Job type61SalaryNo data

Example figures, not a real candidate. Each dimension is scored out of 100 from your own profile, and one we cannot measure says so rather than guessing.