# CipherGate

- **Event:** [Gemini 3 서울 해커톤](https://cerebralvalley.ai/e/gemini-3-seoul-hackathon)
- **When:** Sat, Feb 28 at 9:00 AM – 10:00 PM (GMT+9)
- **Where:** 서울 Seocho District, Olympic-daero, 2085-14 세빛둥둥섬
- **Team:** [Luke Yang](https://cerebralvalley.ai/u/lukeyang)
- **GitHub:** https://github.com/lukeyang/ciphergate
- **Demo video:** https://youtu.be/-_3FCiWS1Ns
- **Gallery:** https://cerebralvalley.ai/e/gemini-3-seoul-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/gemini-3-seoul-hackathon/hackathon/gallery/47

CipherGate | Zero-Knowledge Abuse Screening · Hard Tech / Security

Contact center agents face daily harassment and threats. Traditional moderation processes raw conversations inside screening services, expanding the trust boundary and increasing data exposure risk.

CipherGate asks: Can abuse be detected without reading the message?

Pipeline:
• Gemini 768-dim embeddings generated client-side
• CKKS homomorphic encryption applied to vectors
• Server performs abuse scoring entirely in ciphertext
• Client decrypts score → ALLOW or BLOCK

Plaintext never leaves the client. The screening service never accesses readable content.

Gemini’s high-fidelity embeddings preserve semantic structure even after CKKS encrypted computation, making zero-knowledge policy enforcement viable.

Privacy is enforced by cryptography — not policy.

About: https://ciphergate-customer-app-kkcprzqlcq-uc.a.run.app/about

Demo: https://ciphergate-customer-app-kkcprzqlcq-uc.a.run.app

---

Markdown version of https://cerebralvalley.ai/e/gemini-3-seoul-hackathon/hackathon/gallery/47. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt