01
Multimodal data development
First-person video for model training
- Footage
- Sourced and licensed, including exclusive material
- Rules
- Entity-salience labels and edge cases, co-designed with the science team
- Calibration
- Framework for annotation depth, tested on a sample
- QA
- Production reporting and telemetry per delivery
- The brief
- A video AI lab needed first-person footage and event labels across sports, public safety and news.
- Client context
- Backed by NVIDIA and Amazon · over $200M raised