AI Researcher - Pre-training (f/m/d)

present of AI; Heidelberg; Hybrid; EUR 90,000 - 130,000; Full time

Job summary

Aleph Alpha seeks a researcher to run careful experiments shaping how large foundation models are built and trained, from architecture choices to scaling behavior. A strong candidate has solid math and ML grounding, writes clean PyTorch code, and has trained a language model from scratch at small scale.

Honest reporting of noisy results and fast learning in unfamiliar areas matter more than credentials. This is core model-building work where AI is the product and heavy GPU training is expected.

Qualifications

• Solid math and machine learning grounding.
• Writes clean PyTorch code.
• Trained a language model from scratch at small scale.
• Honest reporting of noisy or failed results and fast learning in unfamiliar areas.

Responsibilities

• Run careful experiments that shape how large foundation models are built and trained.
• Assess architecture and optimizer choices and scaling behavior.
• Monitor experiments and report findings clearly.
• Contribute to reproducible training code and pipelines.

Skills

PyTorch code

Math for ML

Distributed training

Experimentation

Job description

The role centers on running careful experiments that shape how large foundation models are built and trained, from architecture and optimizer choices to scaling behavior. A strong candidate has solid math and machine learning grounding, writes clean PyTorch code, and has trained a language model from scratch at least at small scale. Honest reporting of noisy or failed results and fast learning in unfamiliar areas matter more than credentials alone.

How it fits the AI era

This is core model building work, so AI is the product rather than a tool layered on top. Much routine experimentation and monitoring can be partly automated, but judgment about which experiments matter, reading noisy signals, and deciding what evidence supports stay human. The role rewards people who can reason about why training behaves as it does.

How it builds AI experience

Work here means hands on large scale pre-training on big GPU clusters, with exposure to distributed training, scaling laws, and training stability. That experience is hard to get elsewhere and builds a rare profile. To prepare, reproducing papers at small scale and contributing to open training code are practical ways to show readiness.

Worth showing

• Trained LLM from scratch
• Math for optimization
• Distributed training exposure

This is our own read, written from the employer's posting and the company's Spotlight. The employer's posting is the source of truth for the role.

About Aleph Alpha

Per the company, Aleph Alpha builds sovereign large language models and AI solutions for European enterprises and public institutions.