RAG Chatbot for Docs with Google Drive, n8n, Pinecone & Gemini AI

Developed a Retrieval-Augmented Generation (RAG) based chatbot that allows employees to ask questions and receive instant, AI-generated answers based on internal documents stored in Google Drive.
The system automatically tracks, indexes, and embeds documents using Pinecone and Google Gemini, ensuring answers are always accurate and up-to-date.

---

Companies often struggle to manage and retrieve critical information scattered across multiple internal documents.
This project addresses that challenge by offering:

• A centralized knowledge assistant
• Instant access to policy, technical, or training documents via natural chat
• Automatic document tracking and versioning without manual input

---

• n8n – Automation platform for orchestrating the workflow
• Google Drive API – Triggers on new or updated documents
• Google Gemini (Vertex AI + Gemini-pro) – Used for both text embeddings and generating responses
• Pinecone – Vector database for semantic document search
• Recursive Character Text Splitter – Breaks large documents into manageable chunks
• Window Buffer Memory – Enables multi-turn conversations with short-term context

---

1. Trigger

Listens for new or updated files in a specified Google Drive folder.

2. Document Processing

• Downloads and splits documents
• Embeds text chunks using text-embedding-004
• Stores them in Pinecone for fast vector search

3. User Query Flow

• User asks a question through a chat interface
• Relevant document chunks are retrieved using vector search
• Gemini-pro generates a context-aware, accurate response
• Memory buffer allows follow-up questions for ongoing conversation

---

• Verified with multiple formats: PDFs, Google Docs, and .txt files
• Handled thousands of text chunks with high retrieval precision
• Demonstrated accuracy in retrieving policy information, FAQs, and HR data

---

• Reduces time employees spend searching for information
• Easily scalable across departments or organizations
• Zero manual re-training needed as documents evolve