Sumit Pokharel

• Developing performance-optimized checkout interfaces for 7 critical Ichiba pages in an upcoming revamp, implementing modern accessibility standards for the 50M+ user base.
• Architected and built 20+ reusable React components for the new checkout system, creating a unified component library that will power 200K+ daily transactions post-launch.
• Architected and implemented the complete contactless delivery (okihai) system over 2 months, engineering 7 purpose-built components to optimize no-contact delivery workflows on the order confirmation page.

• Delivered a Python script that streamlines the conversion of digital text into handwritten images and hex code matching for image retrieval from a directory with 1+ million images of handwritten characters.
• Advanced the development of an application to correct Japanese receipt distortions by synthesizing a dataset of artificially altered images & training an advanced open-source model to rectify these distortions.

• Implemented a complete sequence-to-sequence neural network from first principles, faithfully reproducing the seminal Sutskever et al. (2014) architecture for machine translation.
• Built with LSTM encoder-decoder, beam search decoding, CUDA acceleration, and includes pre-trained weights from custom training pipeline of my own.

• Ongoing project replicating foundational ML research papers through complete from-scratch implementations to master transformer architectures and modern LLM designs.
• Implemented the original Transformer, GPT-2, LLaMA-2/3, and Mistral 7B with faithful attention mechanisms and positional encodings, continuously expanding to new architectures.

• Built a reverse-mode automatic differentiation engine from scratch using NumPy, replicating the core computational graph and gradient calculation mechanisms that power PyTorch.
• Implemented dynamic computation graphs, backpropagation algorithms, and gradient accumulation to support fully connected neural network training.

• Developed a command-line AI assistant using Cohere's Command-A model that provides intelligent code assistance directly in the terminal with natural language interaction.
• Built with streaming response handling, advanced markdown rendering, and context-aware workspace analysis that understands project structure and file relationships.

CGPA 3.65