Blender Specialist (AI Computer-Use Trainer)
| Company: | JVLM Tech |
|---|---|
| Salary: | from £22 an hour |
| Hours: | Part-time |
| Working pattern: | Remote |
| Job type: | Contract |
| Posting date: | 30 Aug 2026 |
| Closing date: | 29 Sept 2026 |
Summary
SME Careers partners with leading AI labs to improve how the next generation of intelligent systems learn, reason and act. We are hiring Blender artists to help close a specific gap: current AI agents can read a screen, but they still fail badly at real professional work inside real desktop software, and Blender is one of the hardest cases there is. Dense modal interfaces, keyboard-driven operators, and a 3D viewport where the correct next click depends on the state of the mesh. Agents that handle a web form competently fall apart on a modifier stack.
What you will do:
- Author production-realistic Blender tasks: task briefs reflecting work a studio or client would actually ask for, such as a low-poly prop to spec, a retopology and UV pass, a shader and lighting setup, or a scene assembled to a camera brief, scoped tightly enough that the outcome can be objectively verified
- Record expert demonstrations: set up your own environment with files staged and add-ons in place, then record yourself executing the task end to end with the capture tooling we provide, producing a complete trace of actions, interface state and file changes
- Document the workflow step by step, in a form another expert could follow and reproduce
- Define verification and success criteria: what the finished result must satisfy, which intermediate milestones mark real progress, and which common wrong paths should not be credited
- Cover the range of difficulty, from single-tool operations up to long multi-step workflows that cross applications
- For those who move into QA, audit submitted traces for task realism, recording completeness, step-to-action mapping, and whether the stated success criteria genuinely hold
Defining what "done correctly" means is as important as the work itself. An agent improves only when success can be judged consistently, so expert-authored tasks with expert-defined success criteria are the real bottleneck in training capable agents. Your judgement about what a competent Blender artist would do, and what a plausible-looking but wrong result looks like, is the contribution.
Project standards: recordings must stay free of personal information, only approved applications may be used, and all work must be completed using your own expertise. Use of closed-source AI assistants is contractually prohibited on this project.
There are progression paths into QA and QA lead roles, with ongoing batches as performance and availability align.
Fully remote with weekly pay once work is approved. An interview is required and is conducted in English.