🔥
Nov '25 • General Discussion 💬
The n8n RAG Template Is Incredible. I Extended It With Document Processing So My Agent Can Read Actual Documents. 🔥
Community's 'RAG Chatbot' template: Amazing. But it assumed documents were already text. My client has 200 PDF contracts. Extended template to parse first.
THE BASE TEMPLATE LIMITATION:
Popular RAG templates expect clean text input. Text file → Embed → Store in Qdrant. Perfect IF your documents are already .txt files.
CLIENT'S REALITY:
200 PDF contracts. Scanned images. Multi-column layouts. Tables. Signatures. Stamps. Complex formatting. Can't just 'read file' - need actual parsing.
THE PROBLEM:
Tried feeding PDFs directly to embeddings. Got garbage: partial text, wrong order, missing tables, corrupted formatting.
RAG agent gave terrible answers because vector store contained broken text.
THE EXTENSION (DOCUMENT PREPROCESSING BRANCH):
ORIGINAL TEMPLATE FLOW:
Text File → Recursive Text Splitter → Embeddings Model → Qdrant Vector Store → RAG Agent
EXTENDED FLOW:
Google Drive Trigger (new PDF in folder) → Document Parser → Clean & Structure → Recursive Text Splitter → Embeddings Model → Qdrant Vector Store → RAG Agent
THE NEW NODES:
NODE 1 - GOOGLE DRIVE TRIGGER:
Watches 'Contracts to Process' folder. Triggers on new file.
NODE 2 - DOCUMENT PARSER NODE:
Handles scanned images, complex layouts, tables. Returns clean markdown with proper structure.
NODE 3 - SET NODE (CLEAN & STRUCTURE):
Takes parsed markdown, adds metadata (filename, date, document type), formats for chunking.
NODE 4 - RECURSIVE TEXT SPLITTER:
Same as original template. Chunks cleaned text into optimal sizes for embeddings.
Then connects to existing Qdrant and RAG agent nodes.
WHAT THIS HANDLES NOW:
- Scanned contracts (OCR built in)
- Multi-column PDFs (proper reading order)
- Documents with tables (preserves structure)
- Signatures and stamps (recognized, not noise)
- Word documents (converts automatically)
- Images of documents
THE RESULTS:
- 200 contracts processed: 3 hours (vs 2 weeks manual text extraction)
- RAG agent accuracy: 94% on contract queries
- Handles complex layouts: tables, signatures, stamps
- Vector store quality: Massive improvement
SPECIFIC EXAMPLE:
Query: 'Which contracts have auto-renewal clauses?'
Before extension: Agent couldn't answer (PDFs unreadable)
After extension: Agent found 23 contracts, listed them with clause details
CONFIGURATION DETAILS:
Document parser: Parse mode with LLM enhancement for complex layouts. Returns markdown format.
Metadata: Adds contract_name, signed_date, parties, document_id. Critical for RAG retrieval.
THE LESSON:
RAG templates are incredible. But real documents need preprocessing. Adding document parsing transforms RAG from demo to production.
TEMPLATE:
Complete extended workflow. Document preprocessing branch connects seamlessly to any RAG template. Import, configure folder, deploy.
How did you solve the document format problem in your RAG workflows?
13
17 comments
The n8n RAG Template Is Incredible. I Extended It With Document Processing So My Agent Can Read Actual Documents. 🔥
skool.com/ai-automation-society
Learn to get paid for AI solutions, regardless of your background.
Leaderboard (30-day)
1
+6036
2
🔥
+5184
3
+4460
4
🔥
+2312
5
+830
Powered by