- Own training/eval for VideoClaw’s video models
- Diffusion/flow-matching/video models; shipped end-to-end
- Pedigree-weighted (Meta/ByteDance/Google-type)
- Industry implementer, not pure researcher
VideoClaw is hiring an ML engineer to own the model side of its video harness. The company is starting to train its own small models to make the harness smarter and faster, with proprietary work planned, and this person owns training and evaluation for the models used inside the product. The scope: build benchmarks and feedback loops across video-generation models, and turn product usage and human creative judgment into high-signal training and evaluation data. A core instinct the role demands is knowing what data to collect, label, filter, and feed back to actually improve model behavior.
This hire is explicitly pedigree-weighted, company and specific work matter far more than school. The candidates he was already circling are the shape to target: a Meta ML person, someone involved in training ByteDance’s video models (the “SeeDance”-type models), and someone who trained Google’s video models. They should be industry implementers, not pure researchers: deep research pedigree can be considered, but the person must build and ship, not just advise. Any strong modern generative-media background works: diffusion, flow-matching, video/loop models.
What You’ll Own
- Training and evaluation for the models used inside VideoClaw’s video harness
- Benchmarks and feedback loops across video-generation models
- Turning product usage and human creative judgment into high-signal training and evaluation data
- Proprietary small-model work to make the harness smarter and faster
Requirements
- Hands-on with diffusion, flow-matching, or other modern image and video-generation models
- Has trained, fine-tuned, and shipped useful models end to end
- Strong instinct for what data to collect, label, filter, and feed back to improve model behavior
- Open on the specific ML flavor (supervised, RL, etc.) as long as it is generative-media-related
Background (pedigree-weighted for this hire)
- A recognizable CV name on the CV / video side: Meta, ByteDance (SeeDance-type video models), Google (Veo-type), or comparable frontier video/generative-media labs
Who Will Thrive Here
- A video/generative-media ML engineer from a frontier lab who trains and ships, not just publishes
- Someone with real data instinct for improving model behavior through feedback loops
- Comfortable being the proprietary-model owner at a scrappy team