Synthetic Data Engineer — Diffusion Models

NestAI; Turku; On-site; EUR 90,000 - 130,000; Full time

Job summary

NestAI in Turku, Finland, seeks a seasoned AI researcher to build and operate synthetic data pipelines for object detection in defence contexts. You will train diffusion-based generative models, including LoRA fine-tuning, and combine data sources to improve realism and model performance.

Work closely with the ML Platform and simulation teams to validate data, integrate with infrastructure, and push the boundaries of synthetic data quality for real-world applications.

Qualifications

• Several years of hands-on experience training generative models.
• Experience creating synthetic data in domains with limited real data.
• Deep familiarity with diffusion models and Python ML training frameworks.
• Judgment for when synthetic data is realistic enough to matter for training.
• Comfortable working with imperfect information in a rapidly evolving research area.

Responsibilities

• Build and improve generative pipelines that produce synthetic image and video data for object detection.
• Train and fine-tune diffusion-based models, including LoRA finetuning.
• Combine generative approaches with other data sources to boost realism and variation.
• Validate synthetic data against real-world outcomes to prove training improvement.
• Monitor new open-source models and evaluate their impact on pipelines.
• Collaborate with the ML Platform chapter and simulation team on shared data infrastructure.

Tools

Python

HuggingFace Diffusers

LoRA training

Omniverse

Job description

NestAI in Turku, Finland, seeks a seasoned AI researcher to build and operate synthetic data pipelines for object detection in defence contexts. You will train diffusion-based generative models, including LoRA fine-tuning, and combine data sources to improve realism and model performance.

Work closely with the ML Platform and simulation teams to validate data, integrate with infrastructure, and push the boundaries of synthetic data quality for real-world applications.