WILDFLOW: A Multimodal Retrieval-Augmented Academic Intelligence System

Abstract

Wildflow is a structured multimodal Retrieval-Augmented Generation (RAG) framework designed to enforce strict academic knowledge boundaries in question-answering systems. Conventional RAG architectures reduce hallucination by retrieving relevant documents but still allow large language models to blend retrieved information with internal pretrained knowledge. Wildflow introduces an architectural control layer that prioritizes structured Markdown files as the authoritative source of truth, enforces unit-level vector database isolation, and restricts LLM usage to classification, verification, and bounded reasoning. This approach transforms retrieval from a probabilistic similarity process into a verification-driven academic response pipeline optimized for structured educational environments.