🤷 Nico Martin
banner
nico.dev
🤷 Nico Martin
@nico.dev
Building Transformers.js at @hf.co // #GoogleDevExpert for Web Technologies and AI/ML 💚 #WebAI

https://nico.dev
Is it possible to use the AI SDK with in-browser AI? So on the client side and not on the server side? If so I think we should write an AI SDK adapter for transformers.js..
November 7, 2025 at 11:25 AM
Wait what?! Gemma AND AI sdk? Are you using Gemma 3 on-device / in-browser or in the cloud?
November 5, 2025 at 8:04 PM
www.npmjs.com
October 21, 2025 at 8:34 AM
All in all, it was one of the best conferences I've ever been to! Many thanks to @javve.com and everyone who made it possible.
Oh and yes, it was also my first day at @hf.co. More on that in a separate post.
October 9, 2025 at 12:09 PM
The next day the afterparty started with "Code in the dark". 8 people have 15' to build a website from a screenshot. The catch: they write the html/css but they can never see a preview until the end.
The energy in that room was pretty insane 😅
October 9, 2025 at 12:09 PM
Another highlight was the side activities. On the first evening, you could sign up for “Dinner with 5 new Friends.” You were randomly seated at a table somewhere in Stockholm with 5 other participants. We had a great evening!
October 9, 2025 at 12:09 PM
My talk went great, despite my nerves about presenting it for the first time😅. I was happy, even though a couple demos didn't go exactly as planned, but that's what can happen with live LLM demos. LLMs, which by definition are non-deterministic😄.
October 9, 2025 at 12:09 PM
And then there were the talks. Two days, one stage, excellent speakers, and a schedule that worked perfectly. In terms of topics, there was a wonderful mix of UI, compilers, AI, the Web and general industry topics.
October 9, 2025 at 12:09 PM
The conference and everything surrounding it was unique. So much attention to detail. At the entrance, you were greeted with live music, there was a ball pit, handcrafting tables, sweets, gifts, ... far too much to list everything.
October 9, 2025 at 12:09 PM
Yes, these models are quite large. However, once they have been downloaded, they are stored in the application storage and can be used immediately.
September 30, 2025 at 8:25 AM
It connects to HTTP #MCP servers (like my JokeMCP server) and includes built-in servers like one that captures webcam photos and analyzes them with the #SmolVLM multimodal LLM:

jarvis.nico.dev

github.com/nico-martin/...
September 30, 2025 at 8:00 AM
Tech stack:
- #Qwen3 4B LLM for intelligence
- #Whisper for audio transcription
- #Kokoro for speech synthesis
- #SileroVAD for lightning-fast voice detection
September 30, 2025 at 8:00 AM