FastAI provided me an important base by allowing me to use very high-level libraries to be highly productive in understanding the basic concepts of how to approach Deep Learning and the important techniques. Alas, it didn't get into too much detail. Also, the course was taking a direction which didn't align with my goals. One of my friends, Nishant Nikhil had suggested that to really understand how neural networks work, always implement it from scratch in numpy. He had also taught me Computer Vision Winter Workshop at KGP where we had the same approach. And it worked then, so I was wiling to give it a shot now.
I saw the 3Blue1Brown's Neural Network playlist and covered the first 4 videos which gave a solid intuitive understanding on gradient descent and backprogation.
Then I moved to VikParuchuri/zero_to_gpt to get my hands dirty on the implementation side and implemented gradient descent all by myself. Saw the next video on implementing neural networks but when I went to implement it without looking at the provided solution, I faltered. Tomorrow I am going to watch the explanation video again and try it again without looking at the provided solution.