ArDI: Machine‐Learning‐Driven Raman Phase Analysis for Decoding Complex Mineral Assemblages in Fluid Inclusions

ABSTRACT Fluid and melt inclusions serve as invaluable micro‐scale archives of geochemical processes, preserving snapshots of mineral‐forming fluids and melts across a broad range of pressures and temperatures in natural environments. Their analysis provides critical insights into crystallization histories, metasomatic events, and volatile cycling in geological systems. However, the automated identification of minerals within these inclusions has long been hindered by spectral complexities arising from phase coexistence, host‐mineral interference, and orientation‐dependent Raman signatures. This study introduces the web‐based tool Advanced spectRa Deconvolution Instrument (ArDI) ( https://ardi.fmm.ru ), which integrates fast search algorithms (Faiss), convolutional neural networks, and recursive spectral decomposition to analyze complex Raman spectra. Using pegmatitic quartz as a case study, we demonstrate that the method allows effective identification of minerals within fluid inclusions, e.g., sassolite and ramanite‐(Rb), even when their spectra are integrated with host and other minerals. The developed approach unlocks new possibilities for nondestructive analysis of mineral assemblages in microscopic objects.