Worked on the hpcaitech/TensorRT-Model-Optimizer repository to extend its optimization pipeline with a mixed-precision quantization feature, enabling configurable accuracy and performance trade-offs for deployment in resource-constrained environments. Developed support for INT4 and INT8 quantization strategies, allowing users to specify 8-bit layers through new command-line options. Enhanced the underlying quantization logic by introducing precision mapping and scaling adjustments, which improved inference throughput while maintaining model accuracy. All deliverables were completed and validated with tests, and no major bugs were reported during the period. The work leveraged Python, data processing, and machine learning skills to expand deployment options and business value.