YZ50

My personal work repository for YZ50, a 12-week national talent program for 50 selected young people learning to build, train, evaluate, and improve AI models from scratch.

Weekly Deliverables

Each week consists of individual study from Monday through Thursday and a 60-minute technical discussion on Friday at 17:00. I produce three deliverables:

1. Code → weeks/week-XX-.../code/
2. Experiment notes → weeks/week-XX-.../notes.md
3. 3-slide progress update → weeks/week-XX-.../slides/

Curriculum

Phase 1 · See and Understand (Week 1-4)

Understand how neural networks work before relying on high-level frameworks.

Week	Topic
1	What is a Neural Network
2	micrograd and Automatic Differentiation
3	PyTorch Fundamentals
4	makemore and Language Modeling

Phase 2 · Build and Train (Week 5-8)

Build deeper neural language models and learn to train and debug them reliably.

Week	Topic
5	Embeddings and MLP Language Model
6	Training Mechanics and Debugging
7	Backprop Ninja
8	WaveNet and Deep Architecture

Phase 3 · Transformer, GPT, and Capstone (Week 9-12)

Move on to modern LLM architecture and build a small working GPT model from scratch rather than treating it as a black box.

The program concludes with a mini GPT implementation, an experiment report, and a capstone presentation.

Mentors

Name	Role	Affiliation
Tunga Bayrak	Researcher	NS-AI @ UPenn
Batu El	Researcher	PhD @ Stanford
Mert Cemri	Researcher	PhD @ UC Berkeley
Yavuz Öztürk	Researcher	Founder of Bilkent AI

Repo Layout

yz50/
├── weeks/
│   ├── _template/          # weekly submission template
│   ├── week-01-.../
│   │   ├── code/
│   │   ├── notes.md
│   │   └── slides/
│   └── ...