AARVIK: A Web-Based PDF Question Answering System Using Vector Embeddings and Large Language Models

The huge quantity of data stored in PDF files has made it essential to have smart systems that would help in the first place with automatic information retrieval and comprehension. Searching manually is a laborious process and thus inefficient, and on top of that, it often doesn't even manage to get the semantic context. This paper is about a web-based PDF Question-Answering system called AI-powered Answering with Retrieval, Vectorization, and Interactive Knowledgebase (AARVIK) that was made to fight against this problem. AARVIK utilizes a modern technological stack by merging a Fast API backend and a Stream lit front end. The RAG (Retrieval-Augmented Generation) pipeline is the heart of the system, using Hugging Face's all-MiniLM-L6-v2 Model to create embeddings, FAISS for fast vector indexing, and Google's Gemini 1.5 Flash model for accurate and context-aware answer generation. It has been shown that AARVIK is capable of producing high-accuracy, lowlatency responses, thus significantly outperforming traditional keyword search methods.