Tejasvi

Engineered a transformer model for bilingual translation using PyTorch, incorporating custom tokenization and positional encoding techniques for efficient sequence-to-sequence learning.

Constructed a scalable LMS using Next.js, Prisma, and Postgres, integrating Stripe payments, video on demand via Mux, and dynamic content management with drag-and-drop.

Architected the LLaMA 2 model from scratch in PyTorch, focusing on rotary positional encoding, grouped multihead attention, KV cache, and RMS normalization.

Engineered a scalable chat server using TypeScript, Socket.io, and Kafka for distributed messaging, with PostgreSQL and Redis for robust data persistence.

Implementation of Vision Transformer for image classification, including custom PatchEmbedding layer and complete ViT model tested on Caltech-256 dataset.

Siglip Vision Transformer with spatial attention mechanisms and efficient patch embeddings, featuring multi-headed self-attention and positional encoding.

RESTful API built with Gin framework featuring JWT authentication for secure user login and event management. CRUD operations backed by PostgreSQL.

React Native mobile app for real-time skin disease detection, leveraging a fine-tuned XceptionNet model with 23-class classification on Dermnet dataset.