Computer vision is most of my last year: 4,000 cameras, custom-trained detectors, real-time
tracking. Here is a detector running on your machine, not mine.
CAM // BROWSERidleno signalyolos-tiny · idle
A real object-detection model, running in your browser.
No upload, no server. The weights download once, then it runs on your device.
The same class of model behind intcam's 4,000-camera auditor.
02
Listen
Voice is the interface I keep coming back to, from LiveKit agents to Yarina. Give the page your
microphone and it hears you in real time.
MIC // BROWSERidleweb audio · web speech
Say something. It listens, live.
This page can hear you.
Real-time waveform and speech to text, entirely in the browser.
The primitives behind Yarina, my agent-first voice messenger. Nothing is recorded or sent
anywhere; it never leaves your device.
03
Act
The interesting part is what happens after a system perceives: deciding, then doing. Tell this
agent what you want and watch it drive the page.
agent online. it reads what you type, decides, and acts on this page. try a command, or tap
one below.
A small tool-using agent: intent in, action out. The same loop behind the ops chatbot and
Yarina's v2 agent. The full version swaps these rules for an LLM.
Fourteen years of building things like these, from a fraud service at 120k requests a minute to
a 15-person team to production computer vision. The full record reads on
one page.