ML Engineer, Agents & Reasoning

United States Digital Space LLC; Berlin; Vor Ort; EUR 90.000 - 130.000; Vollzeit

Zusammenfassung

United States Digital Space LLC in Berlin seeks a hands-on ML engineer to build agentic systems for scientific discovery. You will design and deploy systems that reason, plan, and act across experiments, simulations, and datasets, directly interfacing with lab automation.

You will implement planning and uncertainty-aware decision-making, ensure safety and observability, and collaborate with AI researchers to translate models into real-world actions.

Qualifikationen

• 4–8 years hands-on ML engineering experience, preferably with autonomous agents or decision-making systems.
• Experience designing and implementing agent-based systems for real-world workflows including planning, action selection under uncertainty, and stop/fallback/recovery logic.
• Strong track record delivering production-grade ML systems with observability, logging, monitoring, and diagnostics.
• Experience integrating ML/AI models with lab automation, scientific instrumentation, or hardware/software systems.
• Proficiency in Python and at least one major ML framework; strong data tooling (NumPy, SciPy).
• Background in scientific or structured data modeling rather than language-model-first systems.
• Knowledge of safety constraints and safety-aware validation for autonomous decision-making in physical environments.
• Strong cross-functional communication across AI research, engineering, and laboratory teams.
• English fluency; additional languages a plus.
• Right to work in Germany without employer sponsorship — visa sponsorship is not available for this role.

Aufgaben

• Design and implement agentic systems that plan, reason, and act across materials discovery workflows involving experiments, simulations, and scientific datasets.
• Build decision-making systems that select next actions under uncertainty and encode when autonomy should act versus when humans should stay in the loop.
• Implement planning, control logic, and uncertainty-aware decision-making tailored to physical systems and experimental constraints.
• Encode operational, experimental, and safety constraints directly into agent behavior; define stopping criteria, fallback strategies, and recovery mechanisms.
• Collaborate with AI researchers to embed predictive models into agent workflows and translate model outputs into executable real-world actions.
• Integrate agents with laboratory automation and software systems so decisions translate into physical outcomes.
• Instrument agents with logging, monitoring, and diagnostics to ensure observability and support debugging.
• Build evaluation frameworks that assess decision quality, learning efficiency, and overall system behavior.
• Analyze failure cases and iterate on system design based on real-world operational outcomes.
• Own systems end-to-end: from prototype through production deployment and ongoing operation.

Kenntnisse

Python

ML frameworks

Agent systems

Observability

Cross-functional skills

English fluency

Tools

NumPy

SciPy

PyTorch

TensorFlow

JAX

Jobbeschreibung

About the Role

This is a hands-on ML engineering role at the frontier of agentic AI for scientific discovery. You'll build systems that reason, plan, and act inside real materials discovery workflows — turning predictive models into reliable, operational decision-making agents that work directly with physical experiments and laboratory automation. You'll sit at the intersection of AI research, software engineering, and lab science, embedding autonomy, safety, and observability into end-to-end discovery pipelines.

The company is a seed-stage deeptech startup operating in the AI-driven materials acceleration and cleantech space, with a small but highly experienced team and institutional backing. This is an on-site role based in Berlin, Germany. Right to work in Germany without employer visa sponsorship is required.

What You'll Do

• Design and implement agentic systems that plan, reason, and act across materials discovery workflows involving experiments, simulations, and scientific datasets.
• Build decision-making systems that select next actions under uncertainty and encode when autonomy should act versus when humans should stay in the loop.
• Implement planning, control logic, and uncertainty-aware decision-making tailored to physical systems and experimental constraints.
• Encode operational, experimental, and safety constraints directly into agent behavior; define stopping criteria, fallback strategies, and recovery mechanisms to prevent brittle behavior.
• Collaborate with AI researchers to embed predictive models into agent workflows and translate model outputs into executable real-world actions.
• Integrate agents with laboratory automation and software systems so decisions translate into physical outcomes.
• Instrument agents with logging, monitoring, and diagnostics to ensure observability and support debugging.
• Build evaluation frameworks that assess decision quality, learning efficiency, and overall system behavior — going beyond model accuracy alone.
• Analyze failure cases and iterate on system design based on real-world operational outcomes.
• Own systems end-to-end: from prototype through production deployment and ongoing operation.

What We're Looking For

Required:

• 4–8 years of hands-on ML engineering experience, preferably with autonomous agents or decision-making systems in production or applied research settings.
• Demonstrated experience designing and implementing agent-based systems for real-world workflows, including planning, action selection under uncertainty, and defined stopping/fallback/recovery logic.
• Strong track record delivering production-grade ML systems with an emphasis on observability, logging, monitoring, and diagnostics.
• Experience integrating ML/AI models with lab automation, scientific instrumentation, or hardware/software systems.
• Proficiency in Python and at least one major ML framework (e.g., PyTorch, TensorFlow, or JAX), plus strong data tooling skills (NumPy, SciPy, etc.).
• Background in scientific or structured data modeling — rather than language-model-first systems.
• Knowledge of safety constraints and safety-aware validation practices for autonomous decision-making in physical environments.
• Strong cross-functional communication skills; comfortable working across AI research, engineering, and laboratory teams.
• English fluency (additional language skills are a plus).
• Right to work in Germany without employer sponsorship — visa sponsorship is not available for this role.

Nice to Have:

• Experience in materials science, chemistry, or adjacent physical sciences domains.
• Background in probabilistic reasoning, Bayesian optimization, or active learning.
• Familiarity with reinforcement learning, model-based planning, or control theory.Additional European language skills.

Location

This is a full-time, on-site position in Berlin, Germany. Candidates must have the right to work in Germany; visa sponsorship is not provided.