Direct PDF Input for RAG: Exploring New Possibilities with Azure OpenAI

This article explores the exciting potential of using PDF files directly as input for the Azure OpenAI Responses API within a Retrieval-Augmented Generation (RAG) framework, aiming to improve the accuracy of AI-driven responses. The experiment investigates the practical limitations of this approach, specifically considering token usage and processing time when handling different PDF file sizes.

• •The study investigates using the Azure OpenAI Responses API to directly process PDF files as input for RAG.
• •Token count is significantly impacted by the number of characters within a PDF, potentially hitting context window limits for large files.
• •While not suitable for all applications, direct PDF input could improve answer accuracy compared to just using search result chunks in RAG.