The problem zkAd solves

The huge value of targeted advertisement currently leads to ad providers storing and selling all personal & private data. We wanted to solve this problem without compromising on the level of personalized ad quality, and while preserving privacy using zkProofs of LLM-generated ad recommendations.

Challenges you ran into

While we were able to make the prototype work to locally generate the ad recommendation based on the browser history and personal data, we didn’t find any feasible way to compute a zkProof for the LLM inference. We are looking forward to the technology improving further, so that this will easily be possible on local devices.

Technology used

Experimentation with Circom, Llama 3 for ad recommendation, React & FastAPI for the prototype.

Note: Unfortunately our train tomorrow leaves already at 9:30, so we cannot present live. Instead, we’ve recorded the presentation beforehand, and you can find it in the Pitchdeck URL. Thanks for the opportunity 🙏 🚀