Projects

A powerful CLI-based coding agent that helps developers with coding tasks through intelligent tool usage and project management.

in progress

Implemented and trained Qwen3 0.6B from scratch on the fineEDU dataset.

DistilCLIP is a from-scratch implementation of a CLIP-like model, using a Vision Transformer (ViT) as the image encoder and a pretrained DistilBERT as the text encoder. This model was trained on the Naruto BLIP Captions dataset for 25 epochs to understand multimodal representations of anime images and their corresponding captions.

An end-to-end platform empowering Indian farmers with RAG-powered chatbot support, highly accurate crop disease detection, personalized crop recommendations, and real-time commodity price tracking.

A collection of from-scratch implementations of different variants of the attention mechanism.

A from-scratch implementation of GPT-2.