Dr James Ravenscroft on Nostr: Earlier this year I wrote about my handwriting #OCR workflow. I wanted to reduce the ...
Earlier this year I wrote about my handwriting #OCR workflow. I wanted to reduce the friction in this flow so I spent some time building a telegram bot that uses #VLM models to OCR my hand writing. Introducing AnnoMemo which is open source and easyish to self-host. Currently it uses remote models but I'm planning to integrate Qwen2-VL 2B which a) understands my handwriting perfectly and b) runs on my desktop GPU. Considering providing a managed service too
https://brainsteam.co.uk/2024/11/3/03-annomemo-telegram-bot/Published at
2024-11-03 15:18:31Event JSON
{
"id": "0a5db8e0c47ec8619baa568e9c1309f5761eabbb39144558a5b6ef6bc22db643",
"pubkey": "41727036bf8b17d496880125a9ed349c351b8b424384c7314fdfdb2a538b358d",
"created_at": 1730647111,
"kind": 1,
"tags": [
[
"t",
"ocr"
],
[
"t",
"vlm"
],
[
"proxy",
"https://fosstodon.org/users/jamesravey/statuses/113419689115849406",
"activitypub"
]
],
"content": "Earlier this year I wrote about my handwriting #OCR workflow. I wanted to reduce the friction in this flow so I spent some time building a telegram bot that uses #VLM models to OCR my hand writing. Introducing AnnoMemo which is open source and easyish to self-host. Currently it uses remote models but I'm planning to integrate Qwen2-VL 2B which a) understands my handwriting perfectly and b) runs on my desktop GPU. Considering providing a managed service too https://brainsteam.co.uk/2024/11/3/03-annomemo-telegram-bot/",
"sig": "5f37bceaac53d7dae8eb11f8af506017013325248549cf7ac24309276f0742e23c582f8adf7a82393655e7bdcbb71d41cf950417f7ed41c9a38ccde108319fd7"
}