Nick (nt)

the trick to making voice agents fast is pipelining everything: first pipeline: audio packets -> speech to text -> turn-taking model second: LLM -> text to speech -> encoding -> output here's a render of my current latency. bear in mind, I'm running this locally from a wooden hut in the mountains in Turkey - it should be significantly faster once I host it!

• 62 replies
• 8 recasts
• 393 reactions

Here's a mental map 👇 except it's not a map of some idea in your mind, it's a map of your mind and how well it remembered something. i'm using this to predict individual patterns of forgetting during learning in order to optimise how fast I can learn Mandarin.

• 39 replies
• 11 recasts
• 334 reactions

built my own claw - 1.2k LOC (vs. openclaw's 400k and nanoclaw's 7k) - runs claude agent sdk in a container, so it's secure - all the core stuff working: memory, crons and heartbeat - it knows how to install and customise itself (just run claude in the dir) - talk to it on telegram, or just ask it to upgrade itself to use WhatsApp, iMessage, etc, etc https://github.com/NickTikhonov/bitclaw

• 60 replies
• 0 recasts
• 207 reactions

I'm building an app for language nerds and polyglots. I'm using it to learn Mandarin Chinese - it's helping me distill the insights from dozens of research papers on learning into a single, highly effective routine. I'm opening up access to a handful of alpha testers who are learning Chinese, German or Russian. Reply/DM if you're interested.

• 115 replies
• 20 recasts
• 290 reactions