GitHub Repository

Extreme-density data vault for LLM training sets. MsgPack + Zstd-L22 + AES-256-GCM + SHA3-256 Merkle. 39× compression with full encryption and tamper detection.

3 starsPython

Quantum-PULSE – compress-then-encrypt vault for LLM training data

Name: Quantum-PULSE – compress-then-encrypt vault for LLM training data
Availability: InStock
Author: naveenub

by naveenub·Mar 7, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●SolidBig BrainSolve My Problem

Cross-corpus Zstd dictionary + per-record AES-256-GCM, but cryptography audit is missing.

Strengths

•Shared dictionary training across corpus shards means dramatically better compression ratios than vanilla Zstd—genuinely clever approach.
•Merkle tree integrity verification catches silent corruption, meaningful for multi-terabyte training datasets.
•Wire protocol abstraction lets training scripts read vaulted data without decrypting to disk—real security UX gain.

Weaknesses

•No third-party cryptography audit published; claiming QUANTUM-PULSE is auditable while unreviewed undermines the open-source premise.
•Competing directly with Hugging Face Datasets, Unstructured, and commercial ML data platforms—needs to prove dramatically better UX or costs.