By Ayush
This is my personal lab notebook — a quiet record of how I’m learning deep learning, one notebook at a time. I started this to:
- Learn by building, not by binge-watching tutorials.
- Document what works, what breaks, and what I finally understand after a dozen “why is this not converging?” moments.
- Keep a trail of clarity — so when I forget how attention flows or why gradients explode, I can find my way back.
No buzzwords. No over-polish. Just small steps, slow learning, and honest notes.
- From Scratch — Rebuilding concepts like backprop, optimizers, and transformers from the ground up.
- Experiments — CNNs, RNNs, VAEs, transformers, word embeddings — each project started with a what if...
- Reflections — Short thoughts and insights that clicked mid-code.
- Failures that taught more than success — learning to debug exploding gradients, vanishing losses, and wandering logits.
It’s less a repo, more a learning diary — where I try, fail, rethink, and retry.
- No schedule. Add notebooks when curiosity sparks.
- No pressure. Push code when it feels meaningful.
- No pretense. Some notebooks are half-baked — that’s okay. They’re part of the process.
Hey, Future Ayush —
- Remember when the loss curve finally made sense? That moment was earned.
- Revisit the places where you wrote “Why are my gradients exploding again? 😩” — those were real turning points.
- Keep walking toward what feels hard. That’s where growth hides.
With quiet persistence, — Past You
Feel free to explore, borrow ideas, or just scroll through my messy learning process. If something here helps you — even in a small way — I’d love to know. It means a lot.
✨ “Build to understand. Break to learn. Return to grow.”