Andrej Karpathy
San Francisco, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAndrej Karpathy is a machine learning leader and researcher with 15 years of experience building deep learning systems that power real-world products and teaching at scale. He led Tesla’s computer vision team for Autopilot, driving data collection, model training, and deployment on custom hardware toward full self-driving, and was an early research scientist at OpenAI working on generative models and RL. A prolific open-source contributor, he created and refined minimal, educational implementations of transformers and auto-diff (nanoGPT, minGPT, micrograd, char-rnn) that are widely used to teach and prototype modern ML ideas. Comfortable spanning research, production engineering, and developer tools, he pairs deep theoretical understanding with pragmatic optimizations like flash attention, bfloat16 support, and C inference implementations. Based in San Francisco, he also helped popularize deep learning education through Stanford’s CS231n and maintains an influential technical blog that distills complex ideas into accessible demos.
15 years of coding experience
7 years of employment as a software developer
MSc, Computer Science, MSc, Computer Science at The University of British ColumbiaBSc, Computer Science, Physics, BSc, Computer Science, Physics at University of TorontoPhD, Computer Science, PhD, Computer Science at Stanford University
Stackoverflow
1,393reputation
171kreached
13answers
10questions
pdf
top-5%
Github Skills (128)
benchmark10
batch-normalization10
python10
image-processing10
attention-mechanism10
c1110
c1710
lua10
data-loading10
deep-learning10
image-generation10
trainings10
self-attention10
search-api10
multihead-attention10
Programming languages (13)
Github contributions (5)
karpathy/randomfun
Feb 2016 - Dec 2022
Notebooks and various random fun
Role in this project:
ML Engineer
ML EngineerContributions:18 commits, 20 pushes, 1 branch in 6 years 11 months
Contributions summary:Andrej appears to be working on an RNN language model, focusing on the core implementation and experimentation within a Jupyter Notebook environment. They are exploring the process of generating sequences, implementing feedforward passes using basic tensor operations, and evaluating the cross-entropy loss. The code demonstrates a progression through an initial draft with error checks to a functional implementation of forward pass calculations.
pythondata-scienceexplorejupyter-notebooknotebook
karpathy/convnetjs
Apr 2014 - Nov 2016
Deep Learning in Javascript. Train Convolutional Neural Networks (or ordinary ones) in your browser.
Role in this project:ML Engineer
ML EngineerContributions:33 commits, 10 PRs, 10 pushes in 2 years 7 months
Contributions summary:Andrej contributed to the implementation of MagicNet, a system for automated deep learning model selection and ensembling. Their work included the creation of the core `MagicNet` class and associated demo files. The user also extended the `Vol` constructor and optimized a ConvLayer function, contributing to the performance of the deep learning library. Additionally, the user introduced speed tests and refactored some code to improve the library.
convolutional-neural-networksdeep-learningjavascript
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.