Developed 3D Convolution (CONV_3D) support for the TFLite Relax frontend in the apache/tvm repository, enabling 3D convolution operations with configurable padding and fused activations. Extended the operator mapping in tflite_frontend.py and introduced convert_conv3d logic to handle 5D NDHWC layouts, supporting various stride and dilation attributes. Ensured correctness and maintainability by mirroring the established CONV_2D pattern and aligning with TVM issue #19519. Comprehensive unit tests in Python validated functionality across VALID and SAME padding modes and multiple configurations, verifying Relax IR structure. The work leveraged TensorFlow, machine learning concepts, and frontend development skills throughout implementation.