Amir Valizadeh

WORK HISTORY

EDUCATION

ABOUT AMIR VALIZADEH

When GPT-4 launched in 2023, along with every other high school junior at the time, I used it to cheat on my homework assignments. What made me different, though, was that I didn\'t just use it- I became obsessed with understanding how it worked. That curiosity led me to build GPT-2 from scratch, teaching myself transformer architectures and attention mechanisms while still in high school.After this project, I went on to look at internships at major companies and saw things like \"should be able to \'fine-tune\',\'use synthetic data\',\'work with cloud clusters\'\", and so I thought-\"hmm, how can I learn this while proving I learned it.\"- so I created my own end-to-end ML pipeline: generated synthetic training data via OpenAI\'s API, fine-tuned Mistral-7B using Runpod GPUs and Axolotl, and published the results on Hugging Face (700+ downloads, zero marketing).Now I apply these skills across three roles: leading data processing and model training at a University of Pittsburgh ML lab (working alongside grad students), plus an internship and research position. I\'m a firm believer in myself and what I can do with a computer. All I wish for are days longer than 24 hours.