<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>rfbatista</title><link>https://rfbatista.github.io/en/</link><description>Recent content on rfbatista</description><generator>Hugo -- gohugo.io</generator><language>en</language><atom:link href="https://rfbatista.github.io/en/index.xml" rel="self" type="application/rss+xml"/><item><title>About</title><link>https://rfbatista.github.io/en/about/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://rfbatista.github.io/en/about/</guid><description>I&amp;rsquo;m Renan Donato, a software engineer in Brazil. I work on running language models in production, and there&amp;rsquo;s a question that won&amp;rsquo;t leave me alone: what has to be true for high-quality AI to run locally, in Portuguese, on hardware a Brazilian can actually afford?
Everything I publish here comes out of that question. I benchmark Whisper, TTS and wake-word models in Brazilian Portuguese on Raspberry Pi and ESP32, I write about inference (quantization, voice latency, llama.</description></item></channel></rss>