Software engineer, 12+ years building production systems (finance, aviation, public sector, ticketing). On the side I do independent research on efficient inference: quantization, small decision models, and careful benchmarking.

Research and projects

• diffusal: do diffusion LMs tolerate low-bit quantization better than autoregressive ones? Matched pairs, thresholds set before each run. Whitepaper
• newt: retrofitting recurrent depth onto a 4B model, with a noise control that explained away the gains.
• von: a small encoder model that picks between options or scores them in one pass, without generating text. Runs on CPU, Python and TypeScript SDKs.
• rubick: chat, embeddings, and decisions served from one local model.
• vole: a Linux system cleaner, ported from Mole.

Contact: vhugo.dev · Support my open-source work on Ko-fi