after learning about decision trees and classical ml methods, i wanted to go deeper. neural networks always sounded intimidating, but once you break them down, the core idea is surprisingly elegant: a bunch of simple mathematical operations chained together, and somehow they learn to recognize patterns.
this post is my attempt at building one from scratch (well, from keras) to classify handwritten digits using the famous mnist dataset. no fluff, just the essentials.
planning notes: offline vs online, and why deductive planning is cool
i recently started studying planning at college.
i did not expect it to feel this close to the way i think about programming: specify what you want, specify what operations are allowed, then search for a sequence that makes the world look like the goal.
in this post i just want to write down the basics that helped me orient myself.
after weeks of intense work, it’s time to take a step back and relax.
sometimes, the best way to recharge is to disconnect and enjoy the simple pleasures of life. whether it’s reading a good book, or simply spending time with loved ones.
a new chapter begins
the past few months have been transformative.
i successfully defended my thesis and earned my bachelor’s degree, which felt like closing one chapter and immediately opening another. without much pause, i dove straight into my master’s degree in AI while continuing to work. it’s hard to balance both, but the passion for what i’m learning keeps me motivated.