Module 10 Lesson 3: RAG with n8n

AI models only know what they were trained on (the past). RAG (Retrieval-Augmented Generation) allows you to "Inject" your custom data into the AI's memory right when a question is asked.

1. The Three Layers of RAG

1. The Vector Database (The Library): This is where you store your files (e.g., Pinecone, Supabase, or a local Vector store).
2. Embeddings (The Translator): A special AI model that turns words into "Numbers" so the computer can find "Related" ideas.
3. The Retriever: The "Librarian" who goes into the database and finds the relevant pages before giving them to the AI.

---

2. Setting it up in n8n

n8n has a dedicated "Question Answer Chain" node.

• Model: OpenAI.
• Vector Store: Pinecone (or the built-in "In-memory" store for small tests).
• Document Loader: Google Drive or a local PDF.

---

3. The Flow

1. Upload: You upload a 50-page manual.
2. Chunking: n8n breaks the manual into small "Chunks" (paragraphs).
3. Embedding: Each chunk is indexed in the Vector store.
4. Query: A user asks: "How do I reset my password?".
5. Retrieval: n8n finds only the "Password" chunks and tells the AI: "Use THESE 3 paragraphs to answer the user."

---

4. Why use RAG with n8n?

• Privacy: You don't "Retrain" the model. Your data stays in your vector store.
• Accuracy: The AI is less likely to "Hallucinate" because it has the source text right in front of it.

---

Exercise: The Personal Librarian

1. Search for the "PDF Loader" node in n8n.
2. Research: What is "Chunk size" and "Overlap"? (Why do we need them?).
3. Try to build a workflow that reads one PDF and "Answers" a question about it using the In-Memory Vector Store.
4. Why is a Vector Database different from a standard SQL database?
5. Search: What is the "Buffer Memory" node across different n8n chat workflows?

---

Summary

RAG turns an "AI Model" into an "Expert on Your Business." By providing n8n with your private documentation, you build bots that can serve customers, assist employees, and answer technical questions with perfect accuracy and context.

Next Lesson: Extracting value: Summarization and Data Extraction using AI.