My NAS has a load of files dumped onto it over the years. Instead of spending a couple weekends sorting through it like a sane person, I'm making something to automatically index it using the metadata already there.
The goal for marver is to index my NAS and extract as much metadata as it can, then display that in a pretty UI that gives the user multiple ways to refine the metadata, search through their files and make the most of the piled up 1s and 0s they have rotting on a HDD somewhere.
So far I've mostly just been having fun making individual pieces and playing with machine learning (CLIP is really cool!). Right now I'm porting the machine learning code to rust one model at a time so I can bundle the whole app into a single compact docker container, and on the way I get to learn how the black boxes I'm using actually work.
I've tried similar self-hosted apps but they all either have mediocre UI, are painful to setup, or make me add metadata that is already there, just not easily parsed by machines. I want something that most people can deploy in a few minutes and that can get 95% of the way there on its own, while still letting the user refine the results from automated approaches when it inevitably falls flat on its face with edge cases.
It's still far from complete, but my NAS has annoyed me enough recently that I'm starting to focusing on a version that can be deployed, even if it's missing most of the features I want to add. I think once that happens it'll be easier to add features because I can immediately get value from them with, instead of having to slowly struggle towards a functional release.
My NAS has a load of files dumped onto it over the years. Instead of spending a couple weekends sorting through it like a sane person, I'm making something to automatically index it using the metadata already there.
The goal for marver is to index my NAS and extract as much metadata as it can, then display that in a pretty UI that gives the user multiple ways to refine the metadata, search through their files and make the most of the piled up 1s and 0s they have rotting on a HDD somewhere.
So far I've mostly just been having fun making individual pieces and playing with machine learning (CLIP is really cool!). Right now I'm porting the machine learning code to rust one model at a time so I can bundle the whole app into a single compact docker container, and on the way I get to learn how the black boxes I'm using actually work.
I've tried similar self-hosted apps but they all either have mediocre UI, are painful to setup, or make me add metadata that is already there, just not easily parsed by machines. I want something that most people can deploy in a few minutes and that can get 95% of the way there on its own, while still letting the user refine the results from automated approaches when it inevitably falls flat on its face with edge cases.
It's still far from complete, but my NAS has annoyed me enough recently that I'm starting to focusing on a version that can be deployed, even if it's missing most of the features I want to add. I think once that happens it'll be easier to add features because I can immediately get value from them with, instead of having to slowly struggle towards a functional release.