A few open-source projects
People create projects for many reasons. Trying to impress a potential employer, trying to market themselves, etc. The projects below are not designed to impress or persuade, rather, they are the byproduct of my curiosity and obsession over the core components of systems. My objective function is to reduce the complexity of systems and programs, trying to determine if a leaner system, or shorter program exists which fits the criteria of a specific problem.
X/Twitter API - Reverse Engineering
PyPI package with over 1300 GitHub stars, reverse engineering Twitter's undocumented GraphQL, v1.1, and v2 endpoints to perform account actions, search, pull user data, and asynchronously capture large volumes of twitter spaces audio/chat/video data.
Twitter API
Amazon Photos API - Reverse Engineering
I wanted a cheap and reliable place to pull large amounts of data to use to train my models. I noticed there wasn't an API for Amazon Photos, so I made one. Some features include Custom photo organization using PyTorch, FAISS + timm for image similarity search ,efficiently clone local filesystem hierarchy to Amazon Photos ,fast querying (pandas/polars) and metadata storage (parquet) ,exporting your Amazon Photos filesystem hierarchy as tree or JSON ,viewing the filesystem hierarchy
Amazon Photos API
Connectionist Temporal Classification (Paper Implementation)
My implementation of the paper Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks. From scratch, manual gradient calculations, using only NumPy.
CTC
RetinaFace: Single-stage Dense Face Localisation in the Wild (Inference Implementation)
My inference implementation of the paper RetinaFace: Single-stage Dense Face Localisation in the Wild in vanilla JavaScript.
RetinaFaceJS