Hey! So the backstory here is pretty simple. I was going through a rough patch earlier this year and I kept thinking about how weird it is that we have all this technology for connecting people but most of it makes you feel worse not better. And I had this moment where a friend left me a voice note, just a short one, and it genuinely shifted my whole day. Something about hearing someone’s actual voice, the tone and the pauses and all that, it just lands differently than text.
So I built this. You record a voice note between 5 and 60 seconds saying something kind, and it goes into a pool. A stranger somewhere presses play and hears you. Thats it. No accounts, no profiles, no sign up. Completely anonymous.
The moderation side was the interesting engineering problem. Every note gets transcribed by Whisper then a second model classifies it into kind/unsure/unkind/not_a_note. Only “kind” notes ever get served to listeners. I made the decision to run moderation inline during upload rather than batching it, so theres never a window where unmoderated audio is just sitting around waiting. The whole lifecycle gets logged in an append-only event table with request IDs so if something weird slips through I can trace exactly what happend.
Its a pretty standard Next.js + Postgres + S3 setup otherwise. Nothing too clever on the infra side, the interesting stuff is all in the moderation pipeline and figuring out what “kind” actually means when you’re trying to get a model to judge it reliably.
One thing I’m really curious to get feedback on: after you record you get a “ripple link” which is basically your only receipt. You can come back and see how many people heard your note and any anonymous reactions they left. I’m not sure yet if this is compelling enough or if I need to build more of a feedback loop there.
The site is live, you can record or listen right now. Zero friction. Would really love this community’s feedback on the product and the technical approach.
Maybe a middle ground — let people pick a pseudonym and a one-line tagline, but no real identity. You keep the warmth of anonymity while still letting listeners find voices they connected with.
I love this because when I've been down in various periods of my life, after sharing a problem with someone close, one thing they said made all the difference. This has happened to me many times after getting kind words from family and friends. I agree that this will be ruined if it gains traction, but it should be great if it remains not too well known (ironically).
Wow, thanks for the support! To get the first few voice notes, I sent it to family and some colleagues. It was a smoke test to see if anyone would actually record without me literally begging them haha.. and they did! After that, I was confident to start sharing the project more widely and here we are today.
Nice one. I was working on something similar, but then dropped it. The idea was to create a database of diverse voices for AI voice tools to train their models. Of course, it will have user consent. Only those who want to see their voices used in AI models can submit a voice sample.
20 comments
[ 2.9 ms ] story [ 48.2 ms ] threadSo I built this. You record a voice note between 5 and 60 seconds saying something kind, and it goes into a pool. A stranger somewhere presses play and hears you. Thats it. No accounts, no profiles, no sign up. Completely anonymous.
The moderation side was the interesting engineering problem. Every note gets transcribed by Whisper then a second model classifies it into kind/unsure/unkind/not_a_note. Only “kind” notes ever get served to listeners. I made the decision to run moderation inline during upload rather than batching it, so theres never a window where unmoderated audio is just sitting around waiting. The whole lifecycle gets logged in an append-only event table with request IDs so if something weird slips through I can trace exactly what happend.
Its a pretty standard Next.js + Postgres + S3 setup otherwise. Nothing too clever on the infra side, the interesting stuff is all in the moderation pipeline and figuring out what “kind” actually means when you’re trying to get a model to judge it reliably.
One thing I’m really curious to get feedback on: after you record you get a “ripple link” which is basically your only receipt. You can come back and see how many people heard your note and any anonymous reactions they left. I’m not sure yet if this is compelling enough or if I need to build more of a feedback loop there.
The site is live, you can record or listen right now. Zero friction. Would really love this community’s feedback on the product and the technical approach.
See also Mythic Quest's discussion of TTP.[0] User Generated Content is only your friend if you have the resources to moderate it.
[0]https://www.youtube.com/watch?v=3_xqyIMwbew
Https://www.crowdwave.com