We present an integrated framework that leverages large language models to mine microbe–disease prior knowledge, a prior‑guided orthogonal disentanglement network to decompose microbial profiles, and a predictive model for ASD diagnosis using key biomarkers. These three components are now available as ready‑to‑use web applications: LLM‑based Report Generation (https://huggingface.co/spaces/XinyuXu/LLM-based-Report-Generation) for evidence extraction, Microbiome Disentanglement with Prior Knowledge (https://huggingface.co/spaces/XinyuXu/Microbiome_Disentanglement_with_Prior_Knowledge) for interpretable decomposition, and ASD Prediction (https://huggingface.co/spaces/XinyuXu/ASD_Prediction) for clinical classification.