A decoder-only transformer language model implemented from first principles — no high-level model libraries.

The full-stack product powering my startup — from data model to deployment.

A horizontally-scalable inference service with batching, caching, and graceful degradation.

Convolutional neural network for image classification, with a custom training and augmentation pipeline.

A responsive full-stack dashboard with server-side rendering and live charts.

A decoupled, message-queue-based backend designed for resilience and independent scaling.