Local OCR with Ollama Gemma 3 12b · AI Automation Society

Abel Coplet

Mar 14 (edited) • General Discussion 💬

Local OCR with Ollama Gemma 3 12b

Hi, new here. following the content closely.

EDIT. SOLVED

Nevermind, am an idiot, fixed myself, all working correctly.

Leaving it up in case it proves useful to anyone

i am building a document management system, partially based on RAG template by Cole.

one of workflows i need to streamline is fully local OCR. Multiple reasons for that, but one of them is i really like how Gemma 3 model treats the images, and i like the flexibility for doing OCR this way. so with the script i run text extraction from PDFs with gemma 3 locally, and then off it goes to the agent for further processing.

i am a bit of a novice, cant figure out what i do wrong. the Edit fields node correctly recieves the text from HTTP request node as input but Agent not seeing the input. i am sorry for noob question, must be something basic i just cant figure out. this is the output from the fields node [

{

"extractedText": "{{ $json.extracted_text }}"

}

]

also if someone has a better setup for local OCR, let me know

EDIT:

added the following to agent promt {{ $node["Edit Fields"].json.extractedText }}{{ $('HTTP Request').item.json.extracted_text }}

now i get correct desired output

but what am i doing wrong here.

maybe someone with experience can help ? i mean i do get the desired result, but just feels like i am doing something wrong with the set node. Maybe i am missing some edge case scenarios i am not accounting for? i do not necessarily mean the OCR itself, as i have to keep tweaking it for other file type etc, but i mean the framework.

2 comments