home.social

#ishard β€” Public Fediverse posts

Live and recent posts from across the Fediverse tagged #ishard, aggregated by home.social.

  1. HELP ME Fediverse :fediverse: ! πŸ₯Ί I need a name for an open source :opensource: project I'm working on. Please suggest anything in comments! I'm kinda stumped.

    It's a self-hosted AI-enhanced image search engine - see rough design sketches attached.

    ---------

    The Planβ„’:

    Basically it'll index all images in a given directory with 2 mechanisms:

    1. Exif / meta data + filepaths
    2. AI (currently CLIP) embeddings β†’ vector DB

    ....then there'll be a web interface that lets you enter search terms & filter by various things like filename, exif fields, etc

    Quite early in development, but got the embedding worker done today!

    Aim is for it to run comfortably on a Raspberry Pi 4 that also has other things running on it.

    Still to do:

    - 🌐 Web interface
    - πŸŒ„ Exif indexer
    - πŸ₯” Quantise CLIP etc for lower resource usage, go looking for better models
    - :opensource: Open source the code, but need a name for the project first!
    - πŸ’­ Lotsa stuff I forget

    Tech:
    - 🟒 Node.js (pure JS libraries where possible)
    - πŸ‡ VectorDB: npmjs.com/package/hnsw (potential issue: this is in-memory or IndexedDB)
    - πŸŒ„ Exif DB: npmjs.com/package/njodb ('cause I hate working with databases 'cause they're boring - sql ESP)
    - :python: Python worker 4 embedding/AI stuff: PyTorch, CLIP

    #AI #SelfHosting #OpenSource #NamingThings #IsHard #NamingThingsIsHard

  2. HELP ME Fediverse :fediverse: ! πŸ₯Ί I need a name for an open source :opensource: project I'm working on. Please suggest anything in comments! I'm kinda stumped.

    It's a self-hosted AI-enhanced image search engine - see rough design sketches attached.

    ---------

    The Planβ„’:

    Basically it'll index all images in a given directory with 2 mechanisms:

    1. Exif / meta data + filepaths
    2. AI (currently CLIP) embeddings β†’ vector DB

    ....then there'll be a web interface that lets you enter search terms & filter by various things like filename, exif fields, etc

    Quite early in development, but got the embedding worker done today!

    Aim is for it to run comfortably on a Raspberry Pi 4 that also has other things running on it.

    Still to do:

    - 🌐 Web interface
    - πŸŒ„ Exif indexer
    - πŸ₯” Quantise CLIP etc for lower resource usage, go looking for better models
    - :opensource: Open source the code, but need a name for the project first!
    - πŸ’­ Lotsa stuff I forget

    Tech:
    - 🟒 Node.js (pure JS libraries where possible)
    - πŸ‡ VectorDB: npmjs.com/package/hnsw (potential issue: this is in-memory or IndexedDB)
    - πŸŒ„ Exif DB: npmjs.com/package/njodb ('cause I hate working with databases 'cause they're boring - sql ESP)
    - :python: Python worker 4 embedding/AI stuff: PyTorch, CLIP

    #AI #SelfHosting #OpenSource #NamingThings #IsHard #NamingThingsIsHard

  3. HELP ME Fediverse :fediverse: ! πŸ₯Ί I need a name for an open source :opensource: project I'm working on. Please suggest anything in comments! I'm kinda stumped.

    It's a self-hosted AI-enhanced image search engine - see rough design sketches attached.

    ---------

    The Planβ„’:

    Basically it'll index all images in a given directory with 2 mechanisms:

    1. Exif / meta data + filepaths
    2. AI (currently CLIP) embeddings β†’ vector DB

    ....then there'll be a web interface that lets you enter search terms & filter by various things like filename, exif fields, etc

    Quite early in development, but got the embedding worker done today!

    Aim is for it to run comfortably on a Raspberry Pi 4 that also has other things running on it.

    Still to do:

    - 🌐 Web interface
    - πŸŒ„ Exif indexer
    - πŸ₯” Quantise CLIP etc for lower resource usage, go looking for better models
    - :opensource: Open source the code, but need a name for the project first!
    - πŸ’­ Lotsa stuff I forget

    Tech:
    - 🟒 Node.js (pure JS libraries where possible)
    - πŸ‡ VectorDB: npmjs.com/package/hnsw (potential issue: this is in-memory or IndexedDB)
    - πŸŒ„ Exif DB: npmjs.com/package/njodb ('cause I hate working with databases 'cause they're boring - sql ESP)
    - :python: Python worker 4 embedding/AI stuff: PyTorch, CLIP

    #AI #SelfHosting #OpenSource #NamingThings #IsHard #NamingThingsIsHard

  4. HELP ME Fediverse :fediverse: ! πŸ₯Ί I need a name for an open source :opensource: project I'm working on. Please suggest anything in comments! I'm kinda stumped.

    It's a self-hosted AI-enhanced image search engine - see rough design sketches attached.

    ---------

    The Planβ„’:

    Basically it'll index all images in a given directory with 2 mechanisms:

    1. Exif / meta data + filepaths
    2. AI (currently CLIP) embeddings β†’ vector DB

    ....then there'll be a web interface that lets you enter search terms & filter by various things like filename, exif fields, etc

    Quite early in development, but got the embedding worker done today!

    Aim is for it to run comfortably on a Raspberry Pi 4 that also has other things running on it.

    Still to do:

    - 🌐 Web interface
    - πŸŒ„ Exif indexer
    - πŸ₯” Quantise CLIP etc for lower resource usage, go looking for better models
    - :opensource: Open source the code, but need a name for the project first!
    - πŸ’­ Lotsa stuff I forget

    Tech:
    - 🟒 Node.js (pure JS libraries where possible)
    - πŸ‡ VectorDB: npmjs.com/package/hnsw (potential issue: this is in-memory or IndexedDB)
    - πŸŒ„ Exif DB: npmjs.com/package/njodb ('cause I hate working with databases 'cause they're boring - sql ESP)
    - :python: Python worker 4 embedding/AI stuff: PyTorch, CLIP

    #AI #SelfHosting #OpenSource #NamingThings #IsHard #NamingThingsIsHard

  5. HELP ME Fediverse :fediverse: ! πŸ₯Ί I need a name for an open source :opensource: project I'm working on. Please suggest anything in comments! I'm kinda stumped.

    It's a self-hosted AI-enhanced image search engine - see rough design sketches attached.

    ---------

    The Planβ„’:

    Basically it'll index all images in a given directory with 2 mechanisms:

    1. Exif / meta data + filepaths
    2. AI (currently CLIP) embeddings β†’ vector DB

    ....then there'll be a web interface that lets you enter search terms & filter by various things like filename, exif fields, etc

    Quite early in development, but got the embedding worker done today!

    Aim is for it to run comfortably on a Raspberry Pi 4 that also has other things running on it.

    Still to do:

    - 🌐 Web interface
    - πŸŒ„ Exif indexer
    - πŸ₯” Quantise CLIP etc for lower resource usage, go looking for better models
    - :opensource: Open source the code, but need a name for the project first!
    - πŸ’­ Lotsa stuff I forget

    Tech:
    - 🟒 Node.js (pure JS libraries where possible)
    - πŸ‡ VectorDB: npmjs.com/package/hnsw (potential issue: this is in-memory or IndexedDB)
    - πŸŒ„ Exif DB: npmjs.com/package/njodb ('cause I hate working with databases 'cause they're boring - sql ESP)
    - :python: Python worker 4 embedding/AI stuff: PyTorch, CLIP

    #AI #SelfHosting #OpenSource #NamingThings #IsHard #NamingThingsIsHard