Projects

Building and Scaling

Natural Language Processing

project thumbnail

TinyStoriesGPT 5M

A 5 million parameter character level Bigram Transformer implemented in PyTorch and CUDA, trained on the Tiny Stories dataset with context length 128 and batch size 2048, optimized via cross entropy loss for coherent text generation.

project thumbnail

Telugu Automatic Speech Recognition System

Fine-tuned a Wav2Vec2 Large XLSR 53 model on the IIIT Hyderabad ASR Challenge and OpenSLR66 Telugu datasets, achieving a word error rate of 33.44% for robust Telugu speech transcription.

Algorithms & Data Structures

project thumbnail

LeetCode Obsidian Solutions

A curated vault of LeetCode algorithm and data structure solutions formatted for Obsidian, complete with clear explanations and code references to streamline interview prep.

Game Development

project thumbnail

ChessSync

Real time multiplayer chess platform built with Flask and Redis for asynchronous state management, featuring engine validated move logic and cross device synchronization.

Development Tools

project thumbnail

CodeExpert AI Assistant

Jupyter based AI assistant for code queries using LangChain, FAISS vector search, and pre and post processing pipelines to ingest documents and generate natural language responses.