This bot answers questions using Retrieval-Augmented Generation (RAG) and Gemini 3.6 Flash model. It searches the documents you give it, pulls out the most relevant context from your data and so answers stay grounded in your own knowledge base rather than the model's general training. This project serves as a demostration of RAG's most practical usecase which is providing relavent context without the need to re-train the model. For more information of RAG or the project check the github link below.
Step 1: Use the sample document or attach your documents with the paperclip icon in the question box.
Note : The app is a prototype and runs on free tier of vercel and supabase. So, please make sure the documents you are uploading are small (2-3 document, 2-10 Mb and less than 50 pages each) for it to vectorize quickly.
Step 2: The backend vectorizes the document and stores in supabase, the sample documents are already vectorized.Now you can type a question about those documents.
Step 3: The backend retrives relavent context from documents and passes it on to the Gemini-3.6 Flash LLM to answer your query.