A functional document chatbot that eliminates hallucination risk on private content. Any document added to the connected Drive folder is automatically ingested and immediately available for grounded Q&A. Built and demonstrated as a proof of concept.
The Outcome
A two-part system. The first part is a RAG pipeline that watches a Google Drive folder for new files, downloads them automatically, chunks the text, embeds it using OpenAI, and stores the vectors in Pinecone. The second part is a chat interface powered by an AI Agent that retrieves relevant chunks from Pinecone to answer questions. Every answer is grounded in the actual uploaded documents, not model assumptions.
What I Built
Large language models hallucinate when asked questions outside their training data. For use cases involving private documents, internal knowledge bases, or proprietary content, a standard chatbot is unreliable. Every answer needs to be grounded in actual source material.
The Problem
RAG Pipeline and Chatbot