3D Artist - OpenUSD & Synthetic Data · AI/ Computer Vision

About the role

Object-detection models (YOLO) are trained largely on synthetic data: photorealistic 3D scenes of industrial environments—warehouses, production halls, and yards—populated with people, forklifts, machinery, and PPE. A Python-native SimulationApp drives Synthetic Data Generation from an established OpenUSD scene and asset library. The role owns and makes this pipeline dependable by creating new Blender assets, maintaining and extending the USD library, and running SDG batches that deliver clean, training-ready datasets to the ML team on time.

The challenge

The work has downstream consumers—automated pipelines and ML engineers—so broken hierarchies, missing semantic labels, or non-reproducible scenes can cause failed generation runs, wasted GPU hours, and models that miss hazards in the field. Visual quality matters, but consistency, technical correctness, and predictable delivery matter more. The asset library should be treated like production code: versioned, validated, and documented.

What you’ll own

Asset Creation (Blender)

• Model, texture, and optimise new 3D assets: industrial equipment, vehicles such as forklifts and trucks, people, PPE, props, and environmental elements.
• Export assets to USD cleanly and consistently, with correct scale, pivots, hierarchies, naming, materials, and semantic labels required for SDG annotation.
• Keep polycount, texture budgets, and LODs suitable for large-scale batch rendering.

USD Library Stewardship

• Edit, repair, and extend existing OpenUSD assets and scenes, including composition arcs, references, payloads, variants, layering, materials, and semantics.
• Maintain a versioned, documented library with a predictable structure for reliable asset reuse.
• Validate every asset before it enters the library, identifying broken references, incorrect units, and missing semantics before generation runs fail.

Synthetic Data Generation

• Own and operate the end-to-end SDG process: configure and launch generation batches, select scenes, set domain randomisation for lighting, camera poses, textures, and placement, and produce annotation outputs for YOLO training.
• Monitor runs, verify output and label quality, and deliver clean, documented datasets to the ML team on schedule.
• Work with ML engineers to turn model underperformance on specific classes into asset or randomisation changes, then regenerate data.

What you’ll bring

• 3+ years of professional 3D experience in games, VFX, simulation, archviz, or synthetic data, with strong Blender modelling, UV, texturing, and materials skills.
• Hands-on OpenUSD experience, including composition with references, variants, and layers; ability to debug broken stages; and understanding of how incorrect hierarchies or up-axes affect pipelines.
• Working Python knowledge to run, configure, and lightly modify pipeline scripts.
• Computer-vision awareness, including annotation types, domain randomisation, and the sim-to-real gap for detection models.
• High ownership and discipline, including planning work, communicating status honestly, flagging problems early, and reliably completing work independently.
• Communicative English and Polish for daily collaboration.

Nice to have

• Experience with NVIDIA Omniverse, Isaac Sim, or Replicator.
• Experience training or evaluating detection models such as YOLO.
• PBR workflows, Substance Painter/Designer, or photogrammetry.
• Git or other version control for asset management.
• Comfort using AI assistants such as Claude, ChatGPT, or Copilot for scripting, debugging, and learning tools.
• Exposure to industrial, logistics, or safety environments.

What we offer

• A mission with real impact: datasets generated in the role train models that help prevent injuries on industrial sites.
• Full ownership of 3D content and the SDG pipeline, with improvements shipped directly into production models.
• A proven working pipeline rather than a greenfield setup.
• Direct collaboration with the ML team and founders, with short feedback loops.
• Access to frontier AI models as daily support for Python scripting, USD debugging, and Omniverse work.
• Workstation-class GPU hardware and required tools.
• Hybrid work from the Wrocław office.
• B2B contract with compensation matched to experience.

Recruitment process

Selected candidates whose CV and portfolio clearly demonstrate the required skills will be contacted. The process includes a CV and portfolio review, a 5-minute phone call, a 30-minute video interview, and a practical task: model a small asset in Blender, export it to USD according to specification, and explain validation checks. Candidates will receive an update at the end of each stage.

Please include a portfolio or asset samples and, if possible, an example of USD or pipeline work with the application.