my-personal-work

it felt very confusing and error prone whenever i wanted to change yolo models from ultralytics. so i implemented yolo version 11 from scratch, so i can easily customize it.

learning and experimenting with tokenizers for llms (references: andrej karpathy and open resources).

implemented the research paper "accelerating the super-resolution convolutional neural network" from scratch in pytorch.

trained an atari pong agent (pytorch) inspired by karpathy's code using reinforcement learning. compatible with gpu and experiments for cnn / mlp models.

i implemented decoder part of transformer (text generation) in a compact notebook. used the original transformer ideas (decoder-only) to train and sample text.

end-to-end storytelling tool using transformer model and word2vec embeddings; used sentencepiece for tokenization.

a hyper-personalized e-commerce platform, built ai agent using gemini and mistral, vector db with pinecone and faiss, retrieval-augmented-generation, frontend/backend with fastapi, flask, html/css/js.

a toy frontend project to help make favicons and to learn html/css/js more clearly.

toy project to learn basics of html, css and js by implementing the classic rock-paper-scissor game.

automation tool that scrapes data and fills google forms automatically; used example with house prices and addresses from zillow.

nostalgic space invader game built while learning python (inspired by udemy 100 days of python).