metadata
title: README
emoji: 🏢
colorFrom: red
colorTo: indigo
sdk: static
pinned: false
Broadnet Technologies
Enterprise AI that actually works. Built and run on our own infrastructure.
We run our own inference for Arabic and English businesses and publish what we measure, under our own name, with a DOI on every paper. Every model here comes with the paper and the evidence behind its numbers.
- Gemma 4 26B A4B Uno pilot adapter — the first Uno adapter for a mixture-of-experts model, lossless, with its serving profile and measurements on the card. Paper: 10.5281/zenodo.22820511.
- Uno for vLLM — self-drafting speculative decoding with no separate draft model, validated on Ampere, Hopper and Blackwell. Code and container: github.com/brntech/vllm-uno.
- Uno in vLLM (Qwen3-8B), the dense-target study: 10.5281/zenodo.22652609.
- Sensitive-data safety: The Safety Map, 39 models from 14 labs.
Inference API: inference.broadnet.ai · Research: broadnet.ai/research
Also from Broadnet: DialerBee, the AI outbound dialer with answering-machine detection in 11 languages, and broadnet.me, our telecom and omnichannel messaging (bulk SMS, A2P, HLR lookup, SMS firewall).