Show HN: PXG v3.0 – Searchable Compression via 2D Spatial Decomposition

I’ve spent the last year developing a decomposition kernel that addresses the "Linear Bottleneck" in standard compression (Zstd, LZMA, Brotli).

The Problem: Traditional algorithms operate on 1D streams. They are blind to the spatial, temporal, or hierarchical architecture of modern data (3D models, nested JSON, multi-modal blobs). When you force these structures into a 1D window, you leave significant entropy on the table.

The Solution: PXG v3.0 maps binary data to a 13505x13505 grayscale grid. By treating data as a coordinate system rather than a line, our engine utilizes adaptive tiling to isolate redundant blocks across the entire file, not just a sliding window.

Key Benchmarks:

Efficiency: Consistent 60-88% reduction on JSON, PDF, and FBX/3D vertex data.

Searchability: This is an in-place succinct structure. The data is the index. You can query strings, hex patterns, or metadata directly from the compressed binary.

Latency: Seek-latency is clocked at ~0.04ms. Search speed for a complex pose library (225 chunks) is ~6.00ms.

Verification: 100% Byte-Perfect Lossless reconstruction.

The Ecosystem: PXG isn't just a utility; it's designed as the foundation for the AXIS ecosystem—enabling zero-load-time gaming and high-density "Heavy Data" pipelines.

The web demo is live at the URL above. No signups, no trackers. Upload a file (Max 50MB for the demo), download the .pxg, and use the Debug Index to search the compressed results in real-time.

I’ll be here to answer technical questions regarding the tiling logic..