Proficiency in Python with TensorFlow/Keras: model building, fit(), callbacks, and checkpointing
Understanding of CNN architectures, especially ResNet (residual connections, skip layers, BatchNorm) but not just limited to these.
Transfer learning: using pretrained ImageNet weights and fine-tuning for custom datasets
GPU-aware training: mixed precision (tf.float16), tf.data pipelines, multi-GPU strategy on DGX Spark
Data handling: image preprocessing, augmentation (tf.image), tfrecords, tensorflow profiling.
Solid C (not just general C - but embedded C specifically): bitwise operations, uint8t / uint32t, structs, control register manipulation
Digital logic fundamentals - they need to think in hardware terms: clock cycles, control signals, overflow flags and two's complement
Simulation mindset: writing software that models hardware behavior before it runs on silicon