Pawel

DevRel @ai_coustics | Software Engineer | Educator

Winner in the world's largest hackathon (10,000+ projects)

[Pawel @GrungeCoder]

@jack @jack when will it be available to the public?

[Pawel @GrungeCoder]

Voice AI is coming to Berlin!

[ai-coustics @ai_coustics]

Berlin is the capital of sound. 🎙️

It’s home to audio pioneers like @Ableton, Neumann and @SoundCloud, and to a growing community of people building voice and audio products.

On Nov 5, we’re kicking off Voice Mitte: a monthly event for voice and audio builders in Berlin, hosted by ai-coustics, with a different partner company each time. For the first edition we’re teaming up with our neighbors at @tellicom 💙

Expect live demos, a fireside chat with founders from both companies, and time to meet other builders.

Spots are limited, request yours here: https://t.co/JoUEBPGW4D

[Pawel @GrungeCoder]

You may recall my previous demos with Tyto where users where prompted to go to quieter place etc. If you wonder how to build something like this on your own and/or would like to learn how it works behind the scenes - check this out!

[ai-coustics @ai_coustics]

“Sorry, could you repeat that?” is the most expensive sentence in customer support.

Every repeat adds handle time and raises the odds of an escalation. Often the cause is the audio: traffic, a TV in the background, a bad connection. Your agent only finds out once the call has already gone sideways.

Tyto scores audio quality live, during the call, so your agent can react while it’s still happening.

In this walkthrough, our Dev Rel Pawel @GrungeCoder shows how to wire it into your pipeline. It’s for customer support and call centre teams who want to catch audio problems before they hit your satisfaction scores.

Read more in our docs: https://t.co/0J70faUV88

[Pawel @GrungeCoder]

Mentioned "walkman" in the blog, thinking that most likely most of the people would have no idea what that was, especially the younger audience - literally a young male sitting behind me on the train listening something on the walkman! :D

[Pawel @GrungeCoder]

What do slavery, Greek philosophers, 1984 and AI have in common?

After months of silence, I finally forced myself to write some of my thoughts down, even though in the era of AI, doing the thinking yourself gets harder and harder, which is kind of the point of the article.

TV made us watchers. Computers made us builders. Are we slowly becoming watchers again?

Check out my newest blog on my site https://t.co/z8f7w8FC3X

link in the comments.

[Pawel @GrungeCoder]

blog post: https://t.co/iTGB1Hx94K

[Pawel @GrungeCoder]

@BhosalePratim Nice, reminds of my previous apps like Zero-G, Straighty, or lately pullz

Can it do something beyond squats? And how is the form correction? Is it MoveNet model? I find them still way too laggy and inaccurate to provide a proper form feedback.

https://t.co/XP9ckRpeT3

[Pawel @GrungeCoder]

@typesafeai jev is too harsh on my pull ups 😅

Inspired by my previous apps @straighty_app and @0G_app I connected @GoogleDeepMind Gemini 3.8 Live with @typesafeai jev and @ai_coustics Tyto.

Let me know if you would like to give it a try!

[ai-coustics @ai_coustics]

We’re sponsoring #VapiCon2026 by @Vapi_AI!

Nov 11-12, Fort Mason, SF.

Building with voice AI? Come tell us what your audio looks like in production.

🎟️ 20% off with code AICOUSTICS20: https://t.co/NIbsCWMCVz

[ai-coustics @ai_coustics]

Last week the whole ai-coustics team met in Tuscany for our off-site.

We made pasta, biked and hiked, toured a micro winery for a tasting, and went truffle hunting. Somewhere between the trails and the tasting, we also talked a lot about where voice AI goes next.

Voice is turning up in more products every month, and we build the audio layer underneath it.

Alongside the industry, our team is growing, and we’re hiring.

[Pawel @GrungeCoder]

Quite similar feelings here, I only hope we keep it this way! Like the old saying says that hard times create strong people, strong people create good times, and you know the rest of the cycle.

[EP @eptwts]

as a polish person, it's pretty insane to see the narrative flip in the last 10 years...

i was born in poland - at the time it was a country slowly recovering from communism

like millions of other families we were very poor, so we moved to western europe when i was a few years old

for most of my life, to be a polish kid was to be the son of poor migrants - a british person looking up to poland was almost unheard of

in 2020+ the narrative has completely flipped, today somebody moving from poland to the UK is considered an almost unfathomable downgrade

our country has been through a lot - partitions, being the epicenter of both world wars, uprisings, genocide, displacement, communism

it took 30 years of freedom to turn the resilience this has instilled in us + the hunger of being one of the poorer countries in europe, into the country poland is today - a very inspiring success story

the only sad thing about all of this, is that our great grandparents & their parents, the ones who actually endured all of this suffering & fought to make poland what it is today, didn't get to see the country it'd turn into

[Pawel @GrungeCoder]

@dhh you have literally missed Omarchy Warsaw meetup just a couple of days ago 🫠@dhh

https://t.co/efj5CkKwgR

[Pawel @GrungeCoder]

Can robots hang our laundry?

gave this SO-101 arm aka The Raven, one job: hang the sock on the drying rack.

driven by @GoogleDeepMind Gemini 4 Argon

It looks like robots aren't taking over the chores just yet.

[Pawel @GrungeCoder]

@Tesla 19:48 sounds like a case for @ai_coustics Tyto!

[ai-coustics @ai_coustics]

Fireside chats hit different when the fire is real and the view is Tuscany.🔥

During our off-site, we were joined by Sitar Teli from Connect Ventures, one of our lead investors.

Connect is a London-based VC that backs founders at pre-seed and seed, when the idea is still more conviction than proof.

Sitar walked the whole team through how VCs think and what they look for in a company. We wrapped the chat with her favorite sci-fi films. 📽️

Learn more about Connect: https://t.co/Q25WWAusDa

[Pawel @GrungeCoder]

Every time a new model comes out, I'm tempted to give it a robot body and see how it acts. This time it was @GoogleDeepMind Gemini 4 Argon, released just a couple of days ago.

Gemini 4 controls @seeedstudio SO-101 arm. The voice layer runs on Gemini 3.8 Live, and @ai_coustics Tyto tells the arm when someone else's voice cuts into the conversation.

Next step: pairing it with speaker diarization, so the arm also knows who is talking.

I still don't have a proper benchmark template for these experiments. Maybe it's time to standardise one.

The other thing is voice. I always add it because it makes talking to robots feel natural, but it's nearly always a huge workaround. Waiting for the ultimate foundation robotics speech-to-speech model that handles everything!

[ai-coustics @ai_coustics]

@covaldev How to test for it, flag it and fix it, in our joint blog: https://t.co/uZHOdOtzhy

[ai-coustics @ai_coustics]

Which background sounds actually hurt speech-to-text?

Together with @covaldev, we ran 59 call-like clips through 8 streaming STT engines to see which sounds break voice AI in real calls. Noise barely matters until it gets extreme. But interfering speech breaks transcription even when it is much quieter than the caller.

If you only test your voice AI against noise, you are not seeing the full picture.👇

[ai-coustics @ai_coustics]

Machines don't hear like we do.

If someone on a call sounds far away, you notice straight away and ask them to come closer to the phone. You don't think about it.

A voice AI system has no such instinct. It can't tell that something is wrong, and it doesn't know to ask. So it carries on, transcribing words that are slightly off and waiting too long to reply.

If you want your voice AI to understand the audio (and trust us, you do), you should start with audio intelligence to fill this gap. Once a machine can recognize the acoustic conditions, like distance, echo, noise or a second voice, it can respond the way a person would.

This is what we're building at ai-coustics with Tyto. Read more here: https://t.co/cCzPqO85Lj