kanav

your inference setup, explained as a Docker container: - image: frontier model - size: way too big - running: "hello world" RouterPlus right-sizes it → right model per task → up to 80% lower cost → same frontier-level output @useRouterPlus

a model that actually edits instead of regenerating the whole thing

I'M IN! just joined the waitlist for Interface, i genuinely believe this could be gamechanging for a lot of our day to day workflows. you should join it too, use my link to get a headstart: https://natura.inc/i/cepct3

most software engineers will never touch the physical world. Mecka is building custom sensor rigs + CV pipelines that turn messy real-world footage into training data for robots. $60M Series B led by Sequoia. now this is engineering.

TinyFish's growth story has been EXPLOSIVE 🤯 it literally deserves a case study, a lot to learn from them

so you’re telling me you can drop your website into Relay and get: → auto-generated content → 50,000 creators → automatic distribution → performance data feeding back into what you make this is the best time in history to be a founder

every AI voice i've tried, i end up regenerating the same line over and over. routing like openrouter is only half of it. the other half: @OnepinAI checks each line's clarity and pronunciation, then fixes what it got wrong. finally. gonna try this

Benchmarks are cool but I wanted to see what actually happened when I gave 3 STT models the exact same audio. 50 seconds of Steve Jobs' Stanford speech at 1.5x speed. Soniox vs OpenAI vs Google. @soniox_ai came back with a noticeably cleaner transcript, especially when it came