Worked on the google-ai-edge/LiteRT repository to enhance edge AI deployment reliability by both expanding operator support and improving correctness. Addressed a TensorFlow Lite Pack operation bug by ensuring negative axis values are properly adjusted based on output rank, adding new unit tests for robustness. Introduced support for the Qualcomm AI Engine Direct SplitV operator, registering it in the dispatch table and implementing a C API for split management, while optimizing single-split cases to reduce unnecessary computation. Leveraged C++, compiler design, and test-driven development to align LiteRT’s behavior with TFLite specifications, reducing runtime errors and supporting broader model adoption.