Worked on quantized neural network validation and optimization across mozilla/gecko-dev and google/XNNPACK, focusing on test-driven development to ensure correctness and conformance with TFLite. Developed and expanded WebNN quantized operation fusion tests, including dequantize/argMax and GEMM, and improved parameter handling for quantized convolution and convTranspose2d, particularly around padding and zero-point management. Addressed a bug in XNNPACK by correcting INT8 padding-convolution fusion logic and adding tests to verify execution parity. Leveraged C, C++, and JavaScript to implement robust conformance testing, quantization workflows, and performance optimizations, enabling safer, more reliable execution paths for quantized neural network operations.