Sign in

Antonio Zugaldia

@zugaldia.com
143 followers 217 following 103 posts

Building stuff, one thing at a time. Voice engineering at Mapbox. Former WHO and Silica Labs. Physicist. I have a bad feeling about this mission.

PostsRepliesMedia
Antonio Zugaldia @zugaldia.com · 11/04/2026
New Speed of Sound release is out! You can now use NVIDIA Parakeet and Canary models for on-device transcription, alongside Whisper. More models = better accuracy across languages + more offline/private options. Flathub/Snapcraft links & install instructions on www.speedofsound.io/getting-star...
NVIDIA Parakeet and Canary models now available on Speed of Sound.
000
Antonio Zugaldia @zugaldia.com · 04/04/2026
Great news: My little voice typing app for Linux is now available on both Flathub and Snapcraft. Links for both here: www.speedofsound.io/getting-star...
141
Antonio Zugaldia @zugaldia.com · 03/10/2025
Benevolent organizers for life.
020
Antonio Zugaldia @zugaldia.com · 03/10/2025
DevFestDC is today! Find me to talk AI and @mapbox.com (and, oh yes, we're hiring).
110
Antonio Zugaldia @zugaldia.com · 16/09/2025
I'm ready to throw some money at whoever is printing t-shirts with that ultrathink design.
000
Antonio Zugaldia @zugaldia.com · 13/08/2025
Great advice for LLMs in particular and life in general (context: news.ycombinator.com/item?id=4488...).
I've spent a lot of time trying to get LLM to generate things in a specific way, the biggest take away I have is, if you tell it "don't do xyz" it will always have in the back of its mind "do xyz" and any chance it gets it will take to "do xyz". When working on art projects, my trick is to specifically give all feedback constructively, carefully avoiding framing things in terms of the inverse or parts to remove.
100
Antonio Zugaldia @zugaldia.com · 13/08/2025
DevFestDC is back! In collaboration with @developers.google.com, this year we're running a one-day event themed around generative AI and emerging technologies. More information about tickets and how to apply to speak on www.devfestdc.org
010
Antonio Zugaldia @zugaldia.com · 29/07/2025
Clear your LLM context early and often. www.science.org/content/arti...
000
Antonio Zugaldia @zugaldia.com · 05/07/2025
I just posted some notes on how to run this MCP with a local LLM on Linux. Let me know if you run into any issues! github.com/bilelmoussao...
110
Antonio Zugaldia @zugaldia.com · 08/06/2025
I now have Copilot reviewing Claude Code PRs. What a time to be alive.
010
Antonio Zugaldia @zugaldia.com · 30/05/2025
Looks like Anthropic/Claude is giving up on TTS and relying on @elevenlabs.io for their recently announced voice mode. Interestingly, no mention of STT though. Source: app.vanta.com/anthropic/tr...
000
Antonio Zugaldia @zugaldia.com · 09/05/2025
I don't know why I didn't try this before, from screenshot to calendar event in 5 seconds.
000
Antonio Zugaldia @zugaldia.com · 30/04/2025
MCP server idea: Every GTK/GNOME app defines Actions that are available via D-Bus (e.g. screenshot below from GNOME Maps). The MCP server would bridge tool calls with D-Bus calls to any app in the desktop (whether it was designed to be controlled by LLMs or not!).
000
Antonio Zugaldia @zugaldia.com · 28/04/2025
If an image is ~258 tokens, and a token is ~0.75 words, then an image is worth 193.5 words. We've been lied to all our lives.
000
Antonio Zugaldia @zugaldia.com · 15/04/2025
Going with the GSettings approach, it's simple enough and it works. (The goal was to synchronize the application and the extension state, which are two different codebases, one in Python, the other in GJS.)
000
Antonio Zugaldia @zugaldia.com · 21/03/2025
Talk about beginner's luck. My first Bluesky post just got quoted in TechCrunch techcrunch.com/2025/03/21/a... and Techmeme (wow), and in @simonwillison.net's blog (double wow).
070
Antonio Zugaldia @zugaldia.com · 21/03/2025
Looking at their recent Trust Center update (app.vanta.com/anthropic/tr...) looks like they're using Brave under the hood.
172