Worked on the ROCm/tensorflow-upstream repository to enhance the reliability of quantized model conversion by addressing a critical bug in the MLIR-to-TOSA pipeline. Focused on the TosaConvertTFLUInt8 pass, the work introduced robust error handling to abort conversion when encountering tosa.const operations with quantized uint8 values, preventing silent mis-conversions. Expanded and updated test coverage to validate correct handling of ui8 attributes and the new error path, aligning changes with existing stabilization strategies to minimize regressions. Demonstrated proficiency in C++, MLIR, and TensorFlow, with a focus on quantization workflows and automated testing for improved model predictability.