HN Showcase
Back to browse
GitHub Repository

Privacy-preserving semantic search via modular arithmetic. 98% retrieval quality, 8x faster than FHE. Patent pending.

3 starsPython

Zatron – Encrypted semantic search, 98% quality, 8x faster than FHE

by zahraarman·Jun 1, 2026·3 points·0 comments
Visit ProjectView on HN

AI Analysis

●●●BangerWizardryBig BrainZero to One

Encrypted semantic search via modular arithmetic—98% quality, 8x faster than homomorphic encryption.

Strengths
  • •Modular arithmetic transformation obscures embedding structure while preserving 98% retrieval quality
  • •Tested on 350k real Enron emails with correlation ρ=0.034 proving attacker can't recover similarity
  • •Live HuggingFace demo with patent pending shows production-ready implementation not just theory
Weaknesses
  • •Patent pending may limit open-source adoption and community contributions long-term
  • •Top-1 accuracy drops to 55.5% on Enron—near-duplicate emails swap ranks in protected space
Category
Security
Target Audience

Security engineers, privacy-focused developers, enterprises handling sensitive documents

Similar To

Crypten · TenSEAL · Private AI

Similar Projects

Open Source●●Solid

UltraPlot 2.0 – semantic legends, better layouts, faster imports

They rebuilt core legend and colorbar internals into dedicated builders and added a new layout solver — that’s tangible engineering work on the parts of Matplotlib that actually frustrate people. Lazy-loading to cut import time is a nice pragmatic win for large environments, but this is an incremental, well-crafted improvement for Matplotlib users rather than a category‑changing product.

Niche GemWizardry
cvanelteren
305mo ago