zUber is a platform for freelance drivers, focused on privacy-preserving reputation.
The twist here is, zUber combines zero-knowledge & machine learning technologies to enable a new kind of skills' certification system for its users.
We have worked with this idea in mind:
1. You want to prove that you obtained your driver's licence more than 5 years go
2. But you don't want to show it to anyone
1. A picture of your driver's licence goes through a widely-used image recognition model to obtain its date of emission (that's the ML part)
2. At the same time, you generate a zero-knowledge proof that the model ran properly with indeed this date of emission as an output, but your driver's licence image used as the input is kept private (that's the ZK part)
3. You can publicly showcase the date of emission of your driver's licence by veryfing your proof and the date of emission of your licence via a smart contract (that's ZKML, with on-chain verification)
Note: this was a crazy challenge technically! In the end we have only been able to run a simple machine-learning model (handwritten digit recognition).