🧠 PyTorch Deep Learning Foundations (From Scratch)

This repository demonstrates core deep learning engineering skills using PyTorch, built from first principles without relying on high-level abstractions.

The goal of this project is to show that I understand how neural networks actually work, not just how to use pre-built pipelines.

---

🚀 What This Project Covers

This project intentionally focuses on fundamentals that every AI engineer must master:

✅ Custom nn.Module

• Built neural networks by subclassing torch.nn.Module
• Explicitly defined layers and parameters
• Clear separation of model architecture and training logic

✅ Forward & Backward Pass

• Implemented forward pass using tensors and layers
• Used PyTorch autograd to compute gradients automatically
• Explained how gradients flow through the network

✅ Training Loop (From Scratch)

• Manual training loop with:
  • Forward pass
  • Loss computation
  • Backward pass
  • Optimizer step
• No training shortcuts or hidden abstractions

✅ GPU / Device Handling

• Device-agnostic code (CPU ↔ GPU)
• Safe .to(device) usage
• Seamless switching between CPU and CUDA (if available)

✅ Saving & Loading Models

• Saved trained model weights using state_dict
• Reloaded models for inference or continued training
• Demonstrated reproducibility and checkpointing

---

🧩 Project Structure