Javier Boix Campos
ML Engineer
Working where research meets constraints: compressing models, publishing in top journals, and building infrastructure for how science communicates.
About
ML Engineer at Multiverse Computing, where I applied quantum-inspired tensorization to compress a computer vision model from 258 MB to 8 MB. 1.91× throughput, 49% less energy, SSIM 0.97 validated across 20,780 samples. That work is now a commercial selling point for a company that raised +€200 M.
Previously collaborated with UKE Hamburg's Sáez Lab for three years, building CNN pipelines for cell microscopy analysis. This work produced publications in Science Advances and JACC: Basic to Translational Science. Outside of work I build Vivum, an open index for interactive scientific publishing, and serve as Incoming Curator of the Valencia Hub of the Global Shapers Community (World Economic Forum).
Selected Work
-
01
+95% Compression without Accuracy Loss
Applied quantum-inspired tensorization and quantization to compress a proprietary computer vision model by +95%, enabling deployment on edge hardware that previously could not accommodate the model. 1.91× throughput, 49% less energy per inference, SSIM 0.97 on 20,780 samples. Delivered one month ahead of deadline; became a commercial selling point for a company that raised +€200 M.
Read the paper → -
02
Cell Migration Decision-Making
Built CNN pipelines to quantify how cells navigate branching networks, enabling a quantitative study of the trade-off between branching and polarity in migration decisions. Pipelines processed thousands of time-lapse microscopy images, reducing manual analysis time by over 95%.
View in Science Advances → -
03
Vivum
Open index for interactive, web-native scientific articles. Authors self-publish at any public URL and retain full ownership; Vivum provides discovery, community peer review by threshold, and citable DOIs via Zenodo. The "Distill arXiv" that Chris Olah named as missing infrastructure in 2021.
vivum-pub.org →
Journal Publications
- Loading…