Bump! Including markdown to the response output would be great, as it's the standard for models :p
Comment has been collapsed.
Honestly can't really help you with any suggestions either.
Thanks a lot for the huge train though, found some interesting games in there.
Comment has been collapsed.
Did the whole train without bumping, what a crime! Thanks for the train and the AI looks real impressive. I wish I could contribute, but it is much beyond me. I appreciate all the pictures though, it looks fascinating :D
Comment has been collapsed.
Wishing you best of luck with the AI project, I'm totally hopeless with anything more complicated than Windows so I won't try to offer any ideas! Thank you so much for this amazing train!
Comment has been collapsed.
So, a mate of mine starts rambling about this project idea of his — keeping it vague for now 'cause he's gonna try and market it or something — but basically it’s gonna involve AI, voice, the works. Since I’ve been off work recovering from some delightful dental surgery (10/10 don’t recommend), he asked if I could whip up a basic offline AI to help with his prototype.
One week later, in between games and wrangling the kids, I’ve somehow ended up knee-deep in a full-on desktop AI assistant. I’m calling it Version 0.8 for now, with my “MVP” version being 1.0.
Right now it uses FFmpeg, Whisper, LLaMA3, and Coqui TTS. It handles both text and voice input/output, caches WAVs, convos, user settings, and has a few colour themes 'cause who doesn’t love a bit of flair. Currently working on per-conversation caching and trying to make convos reference each other — which is as fun as it sounds.
Also, the AI voice? Sounds like a half-baked call centre operator. Absolutely cooked. I’m adding more voice options soon so it stops sounding like a robo-Karen trying to upsell me internet plans.
Performance-wise, I’ve managed to take voice response from "go make a cuppa" times down to about 6–8 seconds, thanks to streaming chunked WAVs and throwing the GPU at it. Still not lightning, but hey, it’s no longer yelling into the void and waiting for enlightenment.
Anyway, point is — since I was putting together a train anyway, thought I’d ask: anyone got feature ideas? Already blown past what my mate expected, so I’ve got a pretty hefty roadmap going. But I’m all ears for wild suggestions, practical or ridiculous.
Here is your entry to a progressive train. Good Luck and Enjoy ^^
Just finalised the addition of allowing the creation of different conversations, user defined conversation titles, conversational tabbing, persistent / cached conversations and deleting conversations ^^ Currently the entire App is 755 Megabytes. Let's watch that expand >.<
Comment has been collapsed.