Multilingual trend mining tool that discovers early-stage tech ideas from non-English communities and converts them into actionable project insights.

36K parameter MLP that plays Snake using behavioral cloning. Trains in 5 minutes on CPU. No RL, no rewards — just a BFS expert and a classifier.

This project is a deep dive into the architecture of Large Language Models. I built the original Transformer model from the ground up to understand the mathematical foundations of Self-Attention, Positional Encodings, and Layer Normalization. Trained on the Opus Books dataset for machine translation

PyTorch MoE from scratch — gating networks, sparse routing, and expert specialization on MNIST.