Award

Physical Sciences Inc. (PSI) proposes to develop a hardware package for chemical and physical profiling that will provide a common set of measurements for seized drug samples and enable the detection and identification of features useful for tracing drugs back to a common origin. The chemical profiling will be accomplished with any commercial off-the-shelf gas-chromatograph mass spectrometer (GC-MS) system that will be upgraded with advanced algorithms designed to detect co-eluting species and low concentration compounds. The physical profiling will be accomplished using high resolution visible imagery with CNN based identification and computer vision approaches for identifying subtle variations in color, shape, and imprint patterns. A multimodal large language model (MM-LLM) will ingest chemical and physical information and provide descriptions of identified features which will be analyzed with an ensemble classifier to make a final classification of the substance and identify common sources using cluster analysis. Finally, a graph database and search algorithms will ingest physical and chemical information to identify patterns in the manufacturing and distribution of the drugs. The Phase I effort will focus on demonstrating the performance of the chemical and physical profiling algorithms using proxy samples, using the MM-LLM to provide accurate descriptions, and identify common samples. The proposed capability is expected to achieve chemical detection with an 85% probability of detection and identification for co-eluting and low concentration species with a sensitivity of 10 parts per billion range and pill detection with an 85% probability of reidentification from high resolution imagery.