← all repositories
miguelgfierro/ai_projects

A blog in notebook form: one researcher's ML curriculum

A personal collection of tutorial notebooks that double as reference implementations for common ML patterns.

864 stars Jupyter Notebook Learning
ai_projects
Velocity · 7d
+0.2
★ / day
Trend
steady
star history

What it does

This repo is a curated set of Jupyter notebooks covering CNNs, transfer learning, NLP, recommendation systems, fraud detection, and distributed training. Each project pairs working code with a detailed blog post on the author’s site. Think of it as a course syllabus where every lecture is executable.

The interesting bit

The breadth is deliberately eclectic—MXNet and PyTorch sit next to CNTK and fastText, while a t-SNE visualization notebook rubs shoulders with a DeepSpeed distributed training example. The author isn’t chasing framework uniformity; he’s chasing topic coverage. The included ML reference architecture diagram suggests some attempt at systematic thinking amid the sprawl.

Key highlights

  • 12 featured projects spanning computer vision, NLP, time series, and recommender systems
  • Each notebook links to a companion blog post with deeper explanation
  • Includes operationalization examples (Flask APIs, websockets) not just model training
  • One of the few tutorial repos that covers distributed training with DeepSpeed
  • Fraud detection notebook includes GPU-accelerated LightGBM plus real-time serving

Caveats

  • Framework heterogeneity means inconsistent APIs and dependencies across notebooks
  • Some linked libraries (CNTK, MXNet) have declined in community traction since publication
  • No centralized requirements or environment setup; each folder is its own island

Verdict

Good for practitioners who learn by reading code plus prose, and who don’t mind context-switching between PyTorch, Keras, and legacy Microsoft frameworks. Skip if you need a unified, production-ready toolkit—this is explicitly a learning scrapbook, not a product.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.