I need a purpose-built algorithm that aggressively shrinks .glb (glTF binary) files yet can be restored almost instantly in a real-time viewer. File size reduction is the top priority; every byte counts. At the same time the decoder must stay light enough to run on the client side with no noticeable delay, so please keep decompression latency to a bare minimum. You are free to combine or improve on existing ideas—quantisation, re-ordering, mesh prediction, custom binary packing, GPU-friendly streams, etc.—or introduce an entirely new approach, provided the solution remains standalone (no cloud service) and can be distributed under a permissive licence. I will integrate this into both a C++ desktop toolchain and a WebAssembly/JavaScript pipeline, so clean, portable source code is essential. Deliverables • Source code for encoder and decoder, ready to build in C++ (plus optional WebAssembly bindings). • Command-line utility or library interface that accepts a standard .glb and outputs its compressed counterpart, plus a matching decoder. • Test suite with sample assets and a reproducible benchmark report comparing raw .glb size, your output, and decompression time on a mid-range device. • Concise technical write-up explaining the algorithm, its complexity, and any format extensions used. Acceptance criteria (measured on the supplied sample set) • Demonstrated size reduction beyond mainstream tools such as Draco or meshopt when configured for maximum compression. • Decoder initialisation and full restore fast enough for real-time use (single-core execution under typical mobile constraints). • Assets remain visually indistinguishable from the originals at normal viewing distances. If this challenge excites you, let’s discuss sample models and the integration details so you can dive straight into the implementation.