HexaARAI & Web3
- Challenge
- Find participating creators across a large live-streaming platform.
- Idea
- Use visual product recognition to connect streamers with brand support.
- Role
- Technical Lead
- Tech
- Computer vision
- Synthetic training data
- Distributed GPU processing
Mountain Dew Raid used computer vision to find Mountain Dew products in Twitch broadcasts and invite streamers into a creator support programme. The initiative rewarded brand participation rather than an already-large audience.
UNIT9 describes a custom model trained with synthetic imagery, distributed across 50 GPU virtual machines. The system monitored 10,000 streams concurrently and scanned an average of 400,000 unique streamers daily.
- Role
- Technical Lead
- Studio
- UNIT9
- Release
- 2024-01-30
- Case study
- UNIT9 archive
Next Project
View project ↗