Transcribing bank statements to markdown using Gemini Vision AI

by Jimleuk ✓ · 17,188 views · created 2024-09-18 · Document Extraction, Multimodal AI

What this workflow does

This AI workflow has 12 nodes and runs manually, when you click “Execute workflow”. Along the way it connects Google Gemini Chat Model and Google Drive. It also calls stirling-pdf:8080 over HTTP.

How it runs

Trigger: Manually, when you click “Execute workflow”.

1. Get Bank Statement (Google Drive): download
2. Split PDF into Images (HTTP Request): calls stirling-pdf:8080 (POST)
3. Extract Zip File (Compression): uses Compression
4. Images To List (Code): runs custom JavaScript code
5. Sort: sorts the items
6. Resize Images For AI (Edit Image): resize
7. Transcribe to Markdown (Basic LLM Chain): sends a prompt to an LLM, with model: Google Gemini Chat Model
8. Combine All Pages (Aggregate): combines items into one
9. Extract All Deposit Table Rows (Information Extractor): extracts structured fields from text with AI, with model: Google Gemini Chat Model

Services it uses

Google Gemini Chat ModelGoogle Drive

HTTP calls to: stirling-pdf:8080

Before you upgrade to n8n 3.0

Nodes used (11)

• Manual Triggern8n-nodes-base.manualTrigger
• Google Gemini Chat Model@n8n/n8n-nodes-langchain.lmChatGoogleGemini
• Sortn8n-nodes-base.sort
• Google Driven8n-nodes-base.googleDrive
• HTTP Requestn8n-nodes-base.httpRequest
• Compressionn8n-nodes-base.compression
• Coden8n-nodes-base.code
• Edit Imagen8n-nodes-base.editImage
• Aggregaten8n-nodes-base.aggregate
• Information Extractor@n8n/n8n-nodes-langchain.informationExtractor
• Basic LLM Chain@n8n/n8n-nodes-langchain.chainLlm

From the author

This n8n workflow demonstrates an approach to parsing bank statement PDFs with multimodal LLMs as an alternative to traditional OCR. This allows for much more accurate data extraction from the document especially when it comes to tables and complex layouts.…

View the original on n8n.io → Check your own copy